"Oncogenic MAP Kinase Signaling Network" and "Leveraging Social Media Analysis to Inform Tobacco Prevention"

March 30, 2022

Information

Yale Cancer Center Grand Rounds | March 29, 2022

Presentations by: Dr. Benjamin Turk and Dr. Grace Kong

ID7622

To CiteDCA Citation Guide

00:00So it's a couple minutes after the hour,
00:02so why don't we get started?
00:04Welcome to grand rounds.
00:08We have two really interesting talks today.
00:12For those of you who don't know me,
00:16I'll just.
00:18Point out that my name is Ryan Crop.
00:19I'm very new to Yale as the new medical
00:24director for the Clinical Trials Office.
00:27And it's pleasure to start meeting everyone.
00:31So we have two took off.
00:32As I mentioned the 1st.
00:36Benjamin Turk has been
00:39kind enough to join us.
00:40He's an associate professor of
00:42pharmacology and director of
00:44Medical Studies and pharmacology.
00:45He's a member of the CSN program.
00:49His graduate work was in biological
00:52chemistry at MIT and then a postdoc
00:55with Luke Cantley also in Boston,
00:58and he works understanding
01:00molecular mechanisms underlying
01:02signaling pathways and how they're
01:05organized into large networks,
01:07and his lab has been.
01:08Setting protein modifying enzymes
01:11and particularly kinase and
01:13proteases that are important in
01:15in signaling networks and today he
01:18will be talking about the aquaponic
01:21not kinase signaling network.
01:23Thank you Doctor Turk.
01:26OK, thank you. I will share my screen.
01:32See.
01:36OK can you see my screen in the pointer?
01:39Yeah perfect OK great.
01:41Well thank you for that introduction and
01:44also for the invitation to present our
01:48work on MAP kinase signaling networks.
01:51So as we all know,
01:52one of the hallmarks of cancer is
01:55uncontrolled cell proliferation and
01:57survival and cancer cells accomplish this,
02:00at least in part through Co
02:03opting signaling pathways that are
02:05normally activated downstream of
02:08peptide growth factor receptors.
02:10I'm gonna be talking about one of
02:12the major arms of growth factor
02:14signaling the wrasse, RAF, MEK,
02:16Erk, signaling cascade.
02:18So owing to high frequency mutations
02:21in rosanky genes as well as mutation,
02:23amplification of growth factor receptors,
02:26this pathway is amongst the most
02:29highly hyper activated in or more
02:32frequent most frequently hyperactivated
02:34in human cancers.
02:35And though the pathway has been the
02:38subject of intense study for for decades now,
02:41there are still some.
02:42Open questions in the field that our
02:44lab and of course many others are,
02:46are trying to understand and to to
02:48sum up some of these questions that
02:51I'm really talking about today.
02:52One question,
02:53what are the functionally important
02:55components of MAP kinase signaling networks?
02:59So obviously the kinases that form
03:01the core cascade are are have been
03:03well studied or well understood,
03:05but we have less understanding of
03:07other regulators of the pathway.
03:09So for example the protein phosphatases.
03:12That act on these kinases,
03:14and thus attenuate signaling through
03:16the pathway and it in addition.
03:20We don't have a complete catalogue
03:22of the substrates of Erk that act as
03:26the critical effectors in mediating
03:28the cancer cell phone. It type.
03:31So in addition,
03:33one question we're interested in is
03:35how specific connections are made
03:37between the kinases and the regulators
03:39and substrates in this pathway.
03:41So there's been a lot of really beautiful
03:44structural work emerging recently on
03:46the upstream components of the pathway,
03:48in particular,
03:49how Rask connects to RAF and MEK.
03:51But again,
03:52our understanding of the more downstream
03:55components where where we have these
03:57critical effector kinases and their
03:59substrates is is less well understood.
04:01And then lastly,
04:02one thing we know is that the
04:05persistent high level of activation
04:07of this pathway that one gets with
04:11new genic activation really doesn't
04:14faithfully recapitulate the sort
04:16of normal dynamics of activation
04:18when we'd see in response to a
04:21growth factor in a normal cell.
04:23And this can lead to a phenomenon
04:26that someone sometimes called
04:28network rewiring and how.
04:30New or which new connections are
04:32made in these networks and which
04:34connections are broken is is is
04:35something that's important to
04:37know in terms of having a complete
04:39understanding of tumor cell biology.
04:41So I'm I'm gonna tell two stories
04:44briefly today.
04:45The first has to do with the oncogenic
04:48map kinase signaling in in Melanoma.
04:52OK,
04:52so as many of you probably know,
04:55malignant melanomas are really
04:57driven by hyperactive Erk,
04:59MAP kinase signaling and so about half of
05:03melanomas Harbor mutations in the BRAF gene.
05:06Most frequently the V 600 year Leal that
05:09leads to high level constitutive activation.
05:12And the remaining tumors have mutations.
05:15Most of them have mutations either
05:17in the NRAS, GTP ace, the NF One Ras.
05:22GTP is activating protein that negatively
05:25regulates the pathway or gain of
05:28function mutations in in MEK MEK,
05:30one which is just around stream of UVB
05:34graph and and of course the dependence of
05:37melanomas on this pathway has really driven
05:41the development and eventual FDA approval.
05:44Of kinase inhibitors that target both
05:46B RAF and MEK that are currently
05:49used to treat Melanoma,
05:51and while there is a high response rate
05:55for tumors that harbor be RAF mutations,
05:58the the problem with these drugs
06:00and really all targeted therapies is
06:02that the responses are not durable
06:04and patients will relapse within
06:05a few months to a couple of years.
06:09And the most common way that one sees
06:12resistance to these inhibitors is through
06:14re activation of the Erk MAP kinase pathway.
06:17Despite the continued presence of
06:19inhibitor but one can also see
06:21activation of bypass pathways like the
06:23P I3 kinase mtor pathway leading to
06:26resistance and obviously there's been
06:28a lot of interest in understanding
06:30these mechanisms of of tumor cell
06:32resistance to these therapeutic agents.
06:34With the idea that if you understand
06:35how cells become resistant you might
06:37be able to devise.
06:39Addition,
06:39new therapeutic strategies that
06:41might be more durable.
06:43So we got into this area through a
06:46genetic loss of function screen and SH
06:48RNA screen that one of my graduate students,
06:52Eunice Cho conducted to identify
06:54genes that modulate sensitivity to
06:57MEK inhibitors in Melanoma cells and
07:00this work was published last year.
07:03People are interested in getting
07:04more of the details before I talk
07:06about the specifics of this research,
07:08I have to briefly plug the Yale Cancer
07:11Center Functional Genomics core that
07:13I Co direct with David Calderwood.
07:15And really,
07:16the the the the mission of this
07:18core is to facilitate these loss
07:20of function genetic screens.
07:22CRISPR CAS 9 screens or SH RNA screens
07:24like I'm going to talk about and so
07:26hopefully this talk will give you a
07:28flavor of the kinds of information
07:29you can get out of these screens
07:32and inspire you to contact us and
07:34and set up your own.
07:36So I don't have a lot of time to talk
07:37about the details of how the screen works.
07:39Needless to say, we.
07:42In introduce Melanoma cell line
07:45with a pooled SH RNA library.
07:48In this one.
07:48In this case,
07:49one targeting kinases and
07:51phosphatases and then we propagate
07:52in either the presence or absence.
07:54One of two MEK inhibitors tromette
07:56never sell you met in IB and then we
07:59look at which hairpins become enriched
08:01or depleted from the population at the
08:03end of the screen and and what this
08:06will should tell us our our what jeans.
08:09Impact the sensitivity mekan hitters and
08:13hopefully identify additional genetic
08:16modifiers of of map kinase signaling.
08:18So I'm going to jump to the
08:20top hit that came out of this
08:22screen which was a phosphatase.
08:24Assyrian threonine phosphatase called
08:26PPP six seed, and what you can see
08:29here is that amongst all of the
08:30hair pins that were in our library,
08:32those that target PPP succeed were
08:35specifically enriched in the presence
08:37of either Solomon if or trim it nib.
08:39But under control conditions they
08:41were not enriched in the population.
08:42And what that means is that when you
08:44treat cells with a MEK inhibitor,
08:46they grow better if you knock down.
08:49PPP 6C OK.
08:51So seeing PPP 6C as a hit in the
08:54screen really caught our eye and the
08:57reason for that is that something
08:59like 7 to 9% of melanomas have been
09:01shown through genomic analysis.
09:03Whole exome sequencing to harbor
09:06what are thought to be loss
09:08of function mutations in PP6C.
09:10So we thought identifying this
09:12phosphatase in the screen for modulators
09:15of drug sensitivity in Melanoma cell
09:17lines was probably not a coincidence.
09:19So first thing we did was to try to.
09:21Verify this result.
09:23So we derived PDP60 knockout cells
09:26through CRISPR CAS 9 gene editing,
09:28and sure enough,
09:29if we titrate in MEK inhibitor in
09:31this case a trim it and if you can
09:33see that knocking out PPP6C leads to
09:36substantial resistance to the inhibitor
09:38and the other thing that we're observing,
09:40which is kind of interesting,
09:42is that actually sells deleted for
09:45PPP6C grow more poorly than wild type
09:49cell line than the wild type cell line.
09:51But that growth is at growth effect is
09:54actually rescued by low concentrations
09:56of the MEK inhibitor and this is
09:59reminiscent of a phenomenon that's
10:01been seen in in preclinical models,
10:04that's called inhibitor addiction and
10:08basically what what this means is that
10:11it's it's typically characterized
10:13by cells having hyperactive map
10:15kinase signaling and hyperactive
10:17map kinase signaling is toxic to
10:19cells and they can be brought back.
10:21Down into the range that's optimal for
10:24cell growth with low concentrations
10:26of an inhibitor,
10:28and so that was it in a sort of a
10:30immediate clue that of what might
10:32be going on here.
10:34That if loss of PPP6C caused hyper
10:36activation of MAP kinase signaling,
10:39that would explain why you get
10:41resistance because it requires higher
10:43concentrations of drug to suppress the
10:45pathway enough to inhibit cell growth.
10:48And also explain this drug
10:50addiction phenotype.
10:51And sure enough, that's what we see.
10:53So basically,
10:54if we look at a number of
10:57distinct PPP 60 knockout clones,
11:00we can see profound hyperphosphorylation
11:04hyperactivation of of MEK and of Erk
11:06and we can rescue that hyperactivation
11:09by re expressing a wild type allele
11:12of PPP 6C but not a phosphatase dead
11:16allele that's catalytically inactive.
11:18OK,
11:19and we extended these observations
11:20to a whole panel of cell lines.
11:23I'm only showing a few of them here,
11:25basically,
11:26regardless of lineages we lookin cell
11:29lines that either harbor BRAF mutations,
11:32or crass Oren RAST mutations
11:34with a couple of exceptions.
11:37We see that when we knock down PPP
11:4060 by SH RNA, we get increased mech
11:43and or increased ORC phosphorylation.
11:45So we do think this is a general phenomenon,
11:47at least in the context of
11:48oncogenic map kinase. Signaling so.
11:53PPP succeed is a phosphatase and
11:57in experiments that I I I won't
11:59have time to tell you about.
12:01We had ruled out activation of upstream
12:04components of the pathway and had
12:07a good handle on this PB6C acting
12:10at the level of MEK because it's a
12:13phosphatase may most straightforward
12:14explanation would be that it directly
12:16dephosphorylates Mac and we we
12:18do think that's what's going on.
12:20So in in vitro phosphatase assays
12:22we could show that.
12:23Purified PP6P6C complexes.
12:27Candy phosphorylate MEK,
12:28but they don't be phosphorylate Erk,
12:30so there seems to be some substrate
12:32specificity for the upstream component
12:34and probably more compelling we could
12:37detect at least an indirect physical
12:39interaction between Mac and PPP 6C.
12:41So PP6C is the catalytic
12:44subunit of holoenzyme that is,
12:46heterotrimeric,
12:47that includes regulatory subunits
12:49that have ascribed roles and binding
12:52to substrates and recruiting
12:54them for dephosphorylation.
12:55And we could see in komuna precipitation
12:59assays that pulling down any of
13:02the three regulatory subunits.
13:04I will bring down Mac but not so
13:06much with the catalytic subunit,
13:08sort of confirming a role for these
13:11regulatory subunits in in recruiting.
13:13MEC two to the complex.
13:16So I mentioned that PPP 6C is
13:19recurrently mutated in melanomas
13:20and so we wanted to look at whether
13:23these mutations affected signaling
13:24through the MAP kinase pathway.
13:27And so we perform rescue experiments
13:29where we re expressed series of the
13:33the most frequently observed mutants
13:35in our PP60 knockout cells and what we
13:39observed is with a single exception
13:42that these mutants were either
13:45entirely or partially defective.
13:48In their ability to mediate
13:51mech dephosphorylation,
13:52so we conclude that these are
13:55likely partial loss of function
13:57mutations and it sort of makes
14:00sense that they're functioning to
14:03increase signaling through the core
14:05pathway that drives melanomas.
14:08That is, the map kinase signaling pathway.
14:10So, unfortunately,
14:11PPP 60 mutations are are rare enough
14:15that we we really don't know the.
14:18Clinical relevance of these
14:20mutations to pathway activation,
14:23but we were able to mine some data from
14:25C bio portal and it did appear as if
14:28there was a significant correlation
14:31between the M RNA expression level
14:33of PPP6C and the level of either
14:36phospho Erk or Phospho MEK as seen
14:38in reverse phase protein arrays.
14:40So we do believe that PPP 6C is
14:43modulating flux through the pathway in
14:46tumors and and may be a factor that
14:49influences. Therapeutic response.
14:50OK,
14:51so in conclusion of this first part
14:54we've identified PPP 6C as a new
14:57player in restraining oncogenic map
14:59kinase signaling through dephosphorylation
15:00of MEK and that loss of function.
15:03Mutations of PPP 60 lead to hyper
15:06activated Erk signaling some of the
15:08open questions that we're trying to pursue.
15:10Now,
15:11how is PPP 6C regulated?
15:13So this phenomenon where PPP 6C is
15:16required to restrain MEK activation
15:18has has something that we really
15:20only see in the setting of oncogenic
15:22activation of the pathway.
15:23And that suggests to us that maybe
15:26there's a negative
15:27feedback loop where pathway activation
15:29leads to activation of PPP6C
15:32towards the phosphorylation of MEK,
15:34and we'd like to understand how that happens.
15:36And of course, it may be that there
15:38are other signaling outputs substrates
15:40other than mech that are functionally
15:43important for tumors where you see lots
15:45of people pay 60 and we're interested
15:47in trying to identify those as well.
15:50So for the remaining time,
15:51I'm going to switch gears a little bit
15:53and move downstream in the pathway to
15:55do the the the kinase in the bottom
15:58of the map kinase cascade IIRC,
16:01and here the we're going to be
16:03talking a little bit more about the
16:06structural basis for how connections
16:08in the pathway is are made,
16:10and also some of these network rewiring
16:13phenomena they introduced at the
16:14beginning and so the work I'm going
16:16to talk about is the work of really
16:19talented graduate student who's.
16:20Currently in the lab Julissa Torres
16:22Robles and what she was interested in
16:26in looking at our oncogenic mutations in
16:29in Erk 2 itself or encoded by the map K1G.
16:32So as I said at the outset,
16:34you have high frequency mutations in
16:36multiple cancer types of Rasen draft but
16:39at lower frequency you do see mutations
16:41in some of the downstream components.
16:43The Erk mutations in particular are
16:45sort of interesting because you don't
16:47see them in the same tumor types
16:49that you do the Rasen draft mutation.
16:51So where, where as Rasen rap mutations
16:53you you see in melanomas,
16:55lung cancers, colorectal cancers,
16:57pancreatic cancer,
16:58the Erk.
16:592 mutations are largely restricted
17:01to squamous cell carcinomas,
17:03so about 8% of cervical squamous cell
17:06carcinomas have recurrent or two
17:08mutations and about 2% of head and neck.
17:11Squamous cell carcinomas have these
17:13mutations and they've attracted some
17:15attention in that setting because
17:17of potential association between
17:19the presence of those mutations.
17:21And clinical responses to EGF
17:25receptor inhibitors.
17:27So one of the things that kind
17:29of attracted us to this is the
17:31the nature of these mutations.
17:32They're sort of unusual when you compare
17:35them to other activating mutations and
17:37protein kinases that you see in cancer.
17:40So unlike say,
17:41BRAF mutations or EGF receptor mutations,
17:44these mutations don't intrinsically
17:46hyper activate the kinase and they
17:48all map at least in three dimensional
17:51space to a really interesting
17:53region of the kinase catalytic.
17:55So this is a region that falls
17:56outside of the catalytic cleft.
17:58That's known as the common docking group,
18:01and it's called that because
18:02it serves as a hub for protein
18:05protein interactions with ERP two,
18:07so this docking groove binds to
18:10a number of substrates of Erk,
18:13but it also binds to irks regulators,
18:15so the Mach one and Mach 2 which are
18:19the positive regulars that phosphorylate
18:20and turn on or combined at this site,
18:23and the dual specificity phosphatase that
18:26dephosphorylates find it this site as well.
18:29So this sort of presents a little
18:32bit of a conundrum because I just,
18:34you know,
18:34told you that this is a
18:36really functionally important
18:37part of the of the molecule yet,
18:40and so you might expect that mutations
18:42at this site would be loss of function.
18:44But of course just logically it
18:46would seem that mutations in, IIRC,
18:48that you find in cancer should be
18:51gain of function and and the reason
18:53why this is is that these mutations
18:56actually cause selective disruption of
18:58these protein protein interactions.
19:01So for example,
19:03we know that these cancer associated
19:05Earth mutants are still able to
19:07interact with MEK one and MEK two,
19:09and so they can be activated normally,
19:11but they no longer interact with
19:13the dual specificity phosphatase.
19:14So incels, this leads to an imbalance
19:17between their activation and inactivation,
19:19and you accumulate the hyper
19:21phosphorylated active form of the kinase,
19:24but that's not all there is to it because
19:26it turns out that at least one of the
19:29major signaling outputs of Earth that is the.
19:31Chinese risk is also
19:33broken by these mutations,
19:35so these mutants don't interact with
19:36risk and they don't phosphorylate risk,
19:38and so that makes you raised
19:40a few questions in our mind.
19:43So first of all,
19:45what is the scope of interactions
19:47with Erk that are selectively
19:49disrupted by her mutations?
19:51We simply don't know this at this
19:53point and and from a kind of
19:55structural or biochemical standpoint.
19:57Why are some interactions broken
19:59and some spared something that
20:01we we also don't understand?
20:03And so.
20:04In order to address this question,
20:06Jay Lisa conducted a proteome wide
20:09screen to identify sequences that can
20:11interact with the Erk docking group,
20:13and again I don't have time
20:15to explain this in detail.
20:17What we did was mine the human
20:19proteome for short amino acid
20:22stretches of amino acid sequence.
20:24That sort of had sequence similarity
20:27to known interacting sequences like you
20:30would find in in Mach one and Mach 2.
20:34And prepared a genetically encoded
20:35library of about 12,000 sequences.
20:37So these are short sequences,
20:39fragments of proteins that are
20:4114 amino acids long.
20:42And then we use those in a
20:45pooled competitive yeast.
20:46Two hybrid screening format and
20:47and the the bottom line is that you
20:50know similar to sort of an SH RNA or
20:53crisper screen if we have a successful
20:55interaction between Erk and the interactor,
20:58this will become enriched
20:59in the population overtime,
21:00and we can detect this by next
21:02generation sequencing.
21:03So when we do this screen with wild type.
21:05Work we can see that on gratifyingly,
21:09all of the known interactors
21:11interacting fragments that were in
21:13the library actually scores hits.
21:15They become enriched,
21:16and furthermore,
21:17if we align all of these sequences,
21:20we can see a sequence motif.
21:22A signature sequence that emerges
21:23that seems to be a common feature
21:26of sequences that interact with Erk.
21:28So a cluster of proline residues and
21:30a couple of leucine residues close by,
21:33and this is interesting in its own
21:35right because it tells us something about.
21:37How Erk recruits it's interacting proteins,
21:41but what about the mutants? So J.
21:43Lisa conducted this same screen with the two
21:46most recurrent cancer associated mutations,
21:49D321 and E322K and what we saw
21:52was kind of what we expected,
21:54which is that most of the interactions
21:57are preserved about 2/3 of the the
21:59interactors that scored his hits for wild
22:02type or also interact with the mutants,
22:04but about a third of them interacted.
22:07Only with the wild type kinase,
22:09and furthermore,
22:10when we look at the the sequences
22:13that interact only with wild type,
22:16we actually lose this sequence
22:18motif that's characteristic of
22:20of of Erk binders in general.
22:23And actually there's very little
22:25distinguishing feature here,
22:26save for the significant selection of a
22:29single arching residue in the sequence.
22:31So we were a little bit
22:32flummoxed by this at first,
22:34but first we just wanted to
22:36do some basic validation.
22:37I I'm I'm starting to run short on time,
22:40so I'm going to go through this briefly.
22:41Basically,
22:42we could confirm that a
22:43sensually all of the sequences,
22:45but if we if we made synthetic peptides
22:48corresponding to these sequences
22:50that scored as hits in the screen,
22:52we could see that where we expected we
22:56saw differential binding in vitro to
22:58wild type versus mutant alleles of Erk,
23:01one of them in particular peptide
23:03coming from the protein ISG 20 had
23:05particularly high affinity for Erk and
23:07showed the biggest differential binding.
23:09Between wild type and mutant forms.
23:12So we decided to take a structural
23:14biology approach to understand what
23:15was going on here in terms of how
23:17this interacted with her and with a
23:19lot of help from Titus Boggins lab
23:21here in the pharmacology department,
23:23Jay Lisa was able to solve the X-ray
23:25cocrystal structure of wild type work too.
23:27In complex with this fragment of the
23:30ISG 20 protein and I'm just going to
23:32zoom in on the key feature at the
23:35region of ISG 20 that binds to IRK.
23:38That is close to the hot spot for
23:41these mutations we see that the
23:43peptide forms a single turn of an
23:46alpha Helix and that is enforced.
23:48That motive interaction is enforced
23:50by a sequence motif that involves a
23:53hydrophobic isoleucine residue and
23:55then two arginine residues position
23:58close by that actually make direct
24:01polar contacts to the acidic residues
24:04that are mutated in in cancer.
24:07And sure enough, if we then go back.
24:10And look at our sequences.
24:12That bound most preferentially to
24:14wild type the the top 9 sequences
24:16in the original used to hybrid
24:19screening data all have this sequence
24:21motif and we could further confirm
24:25that this motif was important for
24:27binding to wild type IIRC,
24:28but not to mutant forms of work
24:31through in vitro binding assays
24:33that we did with synthetic peptides.
24:35So basically,
24:36if we if we mutate any
24:38of these three residues.
24:40We greatly reduce the binding
24:42affinity with wild type IIRC,
24:44but we have no effect on the
24:46already weak binding affinity
24:48with the mutant forms of FERC,
24:50presumably because the damage had
24:51already been done by those mutants.
24:54So we think we have a good handle on why
24:56some sequences interact specifically
24:57with wild type work and are broken.
25:00The interactions are broken with the mutants,
25:02but we're now trying to do is sort of
25:04understand a little bit more about how
25:06this relates to tumor cell biology,
25:08and this is my last data slide.
25:10And So what we've been doing is looking
25:11at some of the full length proteins
25:14that corresponds to its corresponding
25:15hits from the screen, and one that
25:17in particular that caught our eye,
25:18is the row GTPS exchange factor def.
25:22H1, which has been implicated in a positive
25:25feedback loop for the Erk signaling pathway.
25:29It's a known substrate of of work,
25:31and we can confirm that indietro,
25:33but also also confirm that these
25:36cancer mutated forms of Erk are
25:39unable to phosphorylate.
25:41FH1, at least in vitro,
25:42and we're now following up.
25:45On these studies in head and neck
25:48squamous cell carcinoma cell lines
25:50to see if we can verify this result
25:52and understand what this means
25:54for for tumor cell biology.
25:57So to sum up this part,
26:01we've identified that cancer associated
26:04mutations that map to these common docking
26:08groove of Earth 2 disrupt a subset of
26:12interactions and specifically those
26:14involving a particular sequence motif.
26:16And what we're trying to figure out now,
26:18of course,
26:19is if selective engagement of these
26:21substrates is important for the phenotypic
26:23consequences of work to mutation.
26:25So with that.
26:27I will stop and thank the people
26:29who did the work I mentioned,
26:31Eunice Cho,
26:32who recently left the lab graduated
26:35last year who had done all the work on
26:38PPP 6C and the work on Earth mutants
26:41was conducted by Julissa Torres Robles.
26:44I also like to point out my collaborators,
26:46David Calderwood,
26:47who's my partner in all the
26:50functional genomics stuff.
26:51Tice Boggins lab who helped us with
26:54the crystallography and Mark Gerstein
26:55lab that helped us with the the.
26:57Library design and computational analysis.
27:01And with that I'm happy to take
27:03any questions if we have time.
27:06Thank you that that that was great
27:08and really nice work and and and a
27:11good advertisement for the functional
27:12genomics core 'cause it looks like some
27:15really impressive data we have maybe
27:17two or three minutes for questions.
27:19If you wouldn't mind just putting him in
27:21the chat while people are doing that,
27:24can I just ask you a quick question
27:27about the the PP6C study?
27:31Is it worth you think going back
27:34and trying to redo your your.
27:36Knock down screen in a background
27:40of the the PPP mutant contacts
27:44to see if there's other.
27:46Targets that could restore
27:49sensitivity to the inhibitors.
27:51Yeah, I I do believe so.
27:52And actually one of the things that
27:55we have planned is is such a screen.
27:57So the screen that we did before
27:59was a focus SH RNA library and what
28:01we're gearing up to do is a genome
28:04wide CRISPR screen where we compare
28:06wildtype cells with the PPP 60
28:08knockout cells in the presence or
28:10absence of the of the MEK inhibitor,
28:13and so we're hoping to get out
28:15of that are basically.
28:16We should get genetic modifiers
28:18that affect the growth of the PPP 60
28:21knockout cells and one of the hopes
28:23is that we'll identify potentially
28:26other signaling outputs of PP6C
28:28that are important for growth and
28:29maybe drug sensitivity as well.
28:32You know, it seems like it makes that
28:35make sense in just one other question.
28:38I got a little,
28:39maybe I misunderstood in terms of the.
28:42The prevalence of these mutations
28:44in the in the in that phosphatase,
28:48and they I I thought you had said
28:50that they were relatively common.
28:54It's it's 7 to 9% depending on
28:57the study, so they're they're.
28:59They're not as common it it's.
29:01It's actually interesting
29:01if you look at the data,
29:02they're sort of the I guess
29:04the fifth most common,
29:04you know after the big guys and Ranson.
29:07If one and I think P 53
29:09they they're their next
29:11and do they get enriched? Have you do?
29:13Are there any databases of MEK resistant
29:16MEK inhibitor resistant samples that
29:18you can look to see whether it's
29:20enrichment for that mutation? Yeah,
29:21that hasn't really come out of those studies.
29:23A lot of those studies have been.
29:26Looking at sort of individual
29:29patients and you know people
29:31have made patients right.
29:32Zena graphs and things
29:33like that and and done.
29:34You know whole exome saying there's no.
29:38I mean because they're not
29:40particularly common that it really
29:42has not come up as a bonafide
29:44clinical resistance mechanism.
29:47OK, alright thank. Thank you again.
29:50Really nice work so why don't we
29:53move on to our next presenter?
29:56Is Doctor Grace Kang,
29:59who's an assistant professor in
30:01Department of Psychiatry and a
30:04member of our cancer Prevention
30:06and Control research program.
30:08She did her graduate work in clinical
30:11psychology at Saint Johns and in postdoc
30:14in adolescent addictions in the in the Yale.
30:18A school of Medicine's
30:19division of substance abuse.
30:21Her current research interests
30:23include understanding, substance use,
30:25health disparities among youth,
30:27and the use of social media for
30:29tobacco marketing and and novel
30:31tobacco use behaviors among youth,
30:32and I think she'll be talking
30:35about that today.
30:36Her title is leveraging social media
30:39analysis to inform tobacco prevention.
30:42Dr Kang thank you for for joining us.
30:52And I think you're on mute.
30:57OK, can you hear me now?
30:59Yep perfect OK great thanks.
31:02And you could hear you could
31:03see my slice here, right?
31:04Yeah, OK, awesome, thank you.
31:06Well, thank you so much
31:08for having me here today.
31:10We're gonna really switch gears
31:12and talk about social media
31:14and youth tobacco prevention,
31:16so I will give a brief outline of what
31:20we'll what I will talk about today.
31:25So I'll first given out overview
31:26of why we should care about East
31:29figure prevention in the context
31:30of tobacco prevention and and,
31:32and then the importance of leveraging
31:34social media to understand Easter
31:37youth behaviors and promotion and and
31:39then talk about limitations on current
31:41methods to analyze social media and
31:44then introduce how advances in new
31:46computational methods could be used to.
31:50To overcome some of these limitations,
31:52and then I'm going to talk about
31:54two specific studies in our group
31:56using YouTube data to understand E,
31:58cigarette content and social media.
32:03So cigarette smoking is a leading
32:05cause of preventable cause of death,
32:07disease, disability and death in
32:08the United States and we also know
32:11that smoking causes cancer's of
32:12a variety of charts in the body.
32:15However, cigarette is just one type
32:17of tobacco product in the market.
32:19There are other types of tobacco
32:21products such as cigars,
32:22smokeless tobacco, E cigarettes,
32:23just to name a few, that Berry in harm.
32:27And here what you see is this is
32:30a graph from CDC and this shows.
32:32Different tobacco products and use
32:34rates across the decade and what you
32:37see is overall this decrease in tobacco
32:39use right and but this dotted green
32:42line here is increasing E cigarette
32:44use over the years since 2014,
32:46E cigarettes have been the most commonly
32:49used tobacco product use among youth
32:51and in 2020 more than 4.5 million of
32:54the US youth are are using E cigarettes.
32:58And so when you take E
32:59cigarettes into consideration,
33:00the overall tobacco use rates.
33:02Is increasing among US youth?
33:07So for those who are not that
33:09familiar with E cigarette,
33:10I'll just provide an overview
33:12of what a E cigarette is.
33:14There are many different types
33:15of E cigarettes on the market.
33:17These devices are not regulated,
33:19so there is a rapid innovation such
33:22different product characteristics and E
33:24cigarette devices have evolved overtime.
33:26It first started out with Cigalikes,
33:28which is a which resembles cigarettes.
33:31And then evolve into second
33:33generation on devices like vape pens,
33:35which resembles like a pen.
33:37Third generations are these mods which
33:41vary in how they're it could be really
33:44customized in very different ways,
33:46and it could also excel large
33:49amounts of excelled aerosol,
33:52and then there is this pod mods here
33:54that sort of varies and how it looks.
33:57The most notable device you
33:59may have heard of is Jewel.
34:01They recently got popular because.
34:03They use nicotine salt instead of freebase.
34:05So Freebase nicotine is manipulated
34:07so that it has more of the
34:09harshness or kick the smokers likes.
34:12The nicotine salt is manipulated by
34:14lower the pH level so that it's not
34:17as harsh and allows for higher levels
34:19of nicotine and so the the problem
34:21with using nicotine salt is that
34:24because it's easier to to debate,
34:27you know higher levels of nicotine could
34:30be included in this products and therefore.
34:33You know the initiation among youth
34:35could could be a risk because
34:37of his high level of nicotine.
34:39So once Jewel started really hitting
34:41the market and getting really popular,
34:44this fifth generation of devices
34:46started entering the market and
34:48these are disposable pod devices.
34:50They're meant to be single use
34:51sometimes with multiple packs.
34:53They're small, they're discrete,
34:54they look like jewel they contain.
34:56They also contain they contain salt,
34:58so which has high levels of nicotine
35:00and it comes in multiple flavors.
35:02And there's a widely and importantly,
35:04they're cheap,
35:05so you might see a lot of these products on.
35:08Come in in your gas stations and
35:10other store convenience stores.
35:14So how do you cigarettes work?
35:15You know, even though these
35:17cigarettes vary in how they look,
35:19so the anatomy is is the same.
35:22So it has a component that
35:25holds that you liquid.
35:27It has a heating element.
35:29Any of the power power source in the
35:31form of batteries and is a mouthpiece
35:34in which the user could use to inhale
35:37the aerosol from from the of the
35:39vape and in some in some devices,
35:42just inhaling could activate the device.
35:44So what's in E liquid is made
35:48up of nicotine flavorings.
35:49The base is made up of proper link
35:51like coal and vegetable glycerin,
35:53as well as other additives.
35:54So in terms of nicotine,
35:56that's that's the main drug.
35:58So it stimulates the,
35:59stimulates the central nervous system.
36:01It raises blood pressure,
36:03respiration, heart heart rate,
36:04and releases a feeling of pleasure.
36:07And the the E cigarette that
36:09comes in Freebase comes in zero
36:12to 36 milligrams per milliliter.
36:14The nicotine salt on their
36:16marketed as percentage.
36:17So so for example,
36:19Jewel come as come as 5%,
36:22which is equivalent to about
36:2459 milligrams per milliliter.
36:25And you know the the issue with
36:27labeling is also very important,
36:29because you know 5% of anything
36:31just sounds little right.
36:33But if you actually look at the
36:35milligram per milliliter is actually
36:37very high level of nicotine.
36:38And this is what makes the
36:40nicotine is what makes addictive.
36:41There are zero level of eliquids
36:44and E cigarettes available.
36:46However,
36:47I should say that that's
36:49not that's not very common.
36:51These E cigarettes come in
36:52many different flavors.
36:53There's more than 7000 flavors.
36:56You know it comes in the typical
36:58like menthol tobacco flavor,
36:59but what's really popular or you know,
37:01fruit candy store that desert
37:03kind of flavors.
37:04And also there's also a lot of names
37:07that does not allude to actual,
37:10you know food,
37:11but like obscure names like
37:13you know Unicorn milk,
37:15or you know vampire blood
37:17or things like that.
37:18That gets people's attention.
37:21It is made up of chemicals.
37:24And the people in glycol,
37:25vegetable glycerin and the
37:27combination of the two is used.
37:29The ratio of the two is to create
37:32either more aerosol or less aerosol
37:34is used to intensify flavors or or
37:37a lower the intensity of flavors and
37:40nicotine or other chemicals added
37:42such as other water and other chemicals.
37:44So in addition to you know
37:47nicotine flavor flavorings,
37:48PG,
37:48VG,
37:48and other chemicals E cigarette aerosol
37:51have known or are shown to have
37:54heavy metals volatile organic compounds,
37:55and fine and ultrafine particles
37:57that can be inhaled deeply into the
37:59lungs by both by users as well as bystanders.
38:02The long term effects of this
38:05vaping is currently unknown.
38:07So why?
38:08Why should we care right about E cigarettes?
38:11So nicotine use among youth increases
38:14the risk of lifelong tobacco addiction.
38:17And it could also increase the risk
38:20for future addiction to other drugs as well.
38:22This is this E sticker.
38:24Use is considered an epidemic
38:26in the United States,
38:28so it's NIH, including NCIS.
38:31Research priority priority is to
38:33prevent you thicker E cigarette use.
38:35In fact SCI has RFA specifically
38:38focus on preventing E cigarette use
38:42among youth and has a collaborative.
38:45A grant that's interested in
38:48in particularly interested in E
38:50cigarette preventing E cigarette use,
38:52and then lastly they also have
38:55invested considerable resources into
38:59developingsmokefree.gov,
38:59which has resources to help
39:01youth to quit E cigarette use.
39:04So we're thinking of how
39:05to prevent E cigarette use.
39:07We've got to consider a lot of factors right,
39:09so there are social,
39:11environmental, cognitive,
39:12and genetic influences that plays
39:14a role in in youth tobacco use.
39:17But we also know is that tobacco promotion,
39:19marketing, advertising is causally
39:21related to youth tobacco use and
39:23this has been well established
39:25and has been talked about in
39:27in in surgeon general reports.
39:29So I'm going to focus on social media
39:32because now with the advent of social media,
39:34tobacco promotion really faces a unique
39:37challenge because social media is fast,
39:40it's cheap,
39:40you could reach a lot of people at a quick
39:44speed and it doesn't have sufficient to.
39:46To to control its content.
39:50So it might not be that surprising
39:52to you to hear that you know social
39:55media is popular among youth.
39:5690% of youth have used social media,
39:5975% have at least one active social
40:01media profile and 93% report visiting
40:04on social media site at least daily.
40:07When it comes to understanding how E
40:10cigarettes are promoted to youth is
40:12so important to understand how it's
40:14promoted so pro E cigarette content.
40:16Is on social media through paid ads
40:19through influencers promoting the
40:21products and on post from a share
40:24by their peers and other people?
40:26And recent studies have or are finding
40:28that use of social media among youth
40:31is associated with E cigarette use?
40:34So while there are many different
40:35types of social media platforms in
40:37our in our group or I'm going to
40:39present research findings specific
40:41to YouTube and I'm and I'm sure
40:43all of you have used YouTube so
40:45you're familiar with it.
40:46YouTube is free online streaming service.
40:50Is used by 1.9 billion users,
40:52which is a third of all Internet users
40:54and people spend about a billion hours a
40:57day watching watching online YouTube videos.
40:59So the the data on the right.
41:02The graph here shows this is data from 2018,
41:05so it's a bit old,
41:06but it shows that among teens YouTube
41:09is still popular and actually
41:11there's a recent data that's done.
41:14I think this year last year that showed
41:16that You Tube is still popular among
41:19youth despite newer platforms entering.
41:21That's popular among youth.
41:22We could also see that among those people
41:25who use they they're using YouTube often.
41:31So E cigarettes have been
41:34identified on YouTube.
41:35And people have examined.
41:37Researchers have examined E cigarette
41:39content on YouTube to inform prevention.
41:41They have identified certain
41:42themes that appear in this video,
41:43such as bait tricks that appeal to you and
41:47as well as unorthodox or modify users.
41:50So how people might hack these devices
41:52and use for unintended purposes,
41:54people are examine Instagram videos
41:57to understand whether there's health
41:59warning labels associated with them,
42:00as well as how do these videos explain
42:04health effects of E cigarettes?
42:06And nicotine use as well as
42:09the marketing content.
42:10These are just some examples of what's
42:13been examined on YouTube videos.
42:15However, there is a lot of
42:16limitation in current methods,
42:18so all of these prior studies
42:19have used human coding,
42:21which means that you know we
42:22have humans going in and and and
42:24watching a video to identify these
42:25themes and really limit the number
42:27of videos that could be examined.
42:29So in these studies they examine
42:31about 50 to 350 videos,
42:32but in our previous study we
42:34examined big trip videos on YouTube.
42:36We found that there is like 156,000
42:39videos just on vape tricks along and
42:41other studies have found that 2200
42:43new E cigarette videos are being.
42:45Upload every month.
42:47So,
42:47advances in computational methods
42:49can enhance the methods used to
42:51analyze social media data to
42:53inform tobacco regulatory science.
42:57So the other issue with social
42:59media is that social media custom
43:02tailors the content to users.
43:04So we know that there is a lot of E
43:07cigarette content and this I should
43:09say this algorithm of how social media
43:13content tailors the users is proprietary
43:15and we really don't know what kind
43:18of content user being exposed to,
43:20so understanding the types of content
43:22that you would mute or exposed to is
43:25really important to inform regulations
43:27as well as how to create prevention
43:30strategies such as counter marketing.
43:33And no study has yet.
43:35Try to mimic youth conducting the
43:37search and then apply machine learning
43:38to understand all the data retrieved.
43:44So. So advanced computational
43:47methods can be applied to overcome
43:50these limitations and and gaps,
43:52or another limitation is getting more.
43:54How do we get these data or videos
43:59rapidly so some platforms provide
44:01access via application programming
44:03interfaces APIs while other platforms
44:05require more involved coding to
44:07build data scrapers and API's could
44:10potentially deliver thousands or
44:12even millions of posts per day.
44:14And additionally computational methods.
44:16Can be used to understand topics
44:19related to tobacco prevention
44:20using large social media datasets.
44:22So now I will sort of switch gear to
44:25talk about two studies that we've
44:28used to analyze YouTube content
44:29on E cigarettes and these studies
44:32use unsupervised machine learning
44:33rule based classification,
44:35network analysis as well as
44:38supervised machine learning.
44:39The study one we wanted to understand
44:42whether E cigarette content
44:43on YouTube differs by U2 youth
44:46demographic characteristics.
44:47To understand whether you think content
44:49is being tailored to certain views.
44:51To do this,
44:52we create a 16 fictitious viewer
44:55profiles and these viewer
44:57profiles were separated by age.
44:59So 16 year olds and 24 year olds by
45:02gender as well as race ethnicity.
45:04We may profile for white,
45:05black,
45:05Hispanic youth and we used factory
45:08reset Android phone with Orbot
45:10app to delete all personalization
45:12based on search results.
45:13And these are the search results
45:15are words that we use related
45:17to E cigarettes and we conducted
45:19this search inmate 720.
45:20And we obtain 140 videos which
45:22is equivalent to about 7 pages
45:25of 20 videos per page for each
45:27search word and fix your profile.
45:30And so after we remove all the
45:33duplicates we had 4201 non duplicate
45:35videos in our search result.
45:38The first we wanted to understand,
45:40you know we had to develop a cool bug
45:43to understand what we're examining.
45:45So what we're interested in examining
45:47was like what are the videos being
45:49related to E cigarettes, right?
45:50So were they product reviews,
45:52vape tricks, health information?
45:54You know?
45:55What were these videos talking about?
45:58And then we want to know who are the
45:59people who are uploading these videos,
46:01where they private users,
46:03retailers and we want to know what
46:05types of E cigarette products are
46:07being featured or the eliquids
46:09box mod pods and so on.
46:11We also want to see if there were
46:13actually selling these products
46:15to youth and so we buy we look to
46:17see whether this external links
46:19for purchasing and discount codes.
46:22So once we quoted this book, I'll catbug.
46:24We're two independent reviewers
46:26randomly review the finalizer themes,
46:28and then we establish integrative
46:31reliability.
46:32And then after that one quarter
46:34labeled 1000 videos,
46:36which was used to train supervised
46:38machine learning algorithms for study one,
46:41I'm going to focus on video themes
46:43because our goal was to understand
46:44whether the video theme content
46:46was different among users.
46:48However,
46:48the methods are the same for both studies.
46:52So using network analysis we plotted
46:54exposure similarities as a network of
46:56demographic attributes and videos.
46:58So what you see here is a graph of male,
47:00female and by different age groups and
47:02the thickness of this purple line indicate
47:05the normal number of common videos.
47:07So what we see that both 24 year old
47:10profiles have the most most videos
47:13in common and then it's 24 year
47:16old male and 16 year old female.
47:18And we also use K means clustering,
47:20which is a powerful unsupervised machine
47:23learning algorithm that finds similarity
47:25between items and grouped them into
47:27clusters without the human input.
47:29And then we used human data.
47:33A human labeled data as an input to
47:36graph convolutional network for machine
47:39based classification of the 4201 videos,
47:42titles and descriptions.
47:44And we found that just north of high
47:47accuracy and using GCN we were able to
47:51identify what the video themes were.
47:55So 49% of the videos were product reviews,
47:5926.9 videos.
48:00Or informational or or modifying
48:01so these are videos that teaches
48:04people how to use an E cigarette or
48:06how to modify or hack in E cigarette
48:0815% or health information.
48:10Videos about E cigarettes and 9% were
48:14just like other types of videos.
48:19And so after performing clustering
48:21classification, we calculate the
48:23percentage of each video type in
48:25each category by demographic groups.
48:28So what we find here is that.
48:32The green color is the product of you,
48:34so these are videos that talk about you
48:36know like give product reviews on the
48:38product and we find that the product
48:39reviews represented by the green color
48:41is more common among 24 year old profiles.
48:45Health health is represented by
48:47Orange is similar or cross a little
48:50bit more common among males.
48:53And what you what's interesting
48:54here is that the lighter bluish
48:56purplish color here is informational
48:58videos where how to use an Instagram
49:01or how to modify an Instagram.
49:03And that's a lot more common
49:05among underage female group.
49:07And other other videos are more common,
49:10represented by the darker purple
49:12here for male 16 year olds,
49:14which is concerning because these
49:16videos had content like you know
49:18related to cannabis vaping and
49:20other vape tricks and so on.
49:22So there is concerning content
49:24that shows that more tailored
49:26towards younger younger youth.
49:28So our results show that demographic
49:31attributes does factor into
49:33YouTube algorithmic systems.
49:35In the context of esseker
49:37related queries on YouTube,
49:38we found that the similarities between
49:41exposure for male and female 24 year
49:43olds and actually higher than than
49:46the connection between other pairs.
49:47We also found that underage
49:49users work more exposed to more
49:51instructional videos on E cigarettes,
49:53while all the age groups were
49:55most exposed to product reviews.
49:57So all of this is concerning because.
50:00We because this shows that underage profiles,
50:04right so 16 year olds are able to or
50:06are exposed to E cigarette content
50:09despite YouTube having policies
50:12about prohibiting Easter great
50:14content to their underage viewers,
50:16such as product reviews.
50:20So now I'll talk about our second study.
50:24So we identify we have four areas
50:26of interest, which is, you know,
50:27what are the video themes?
50:28Who are the people uploading these videos?
50:31You know what types of E cigarette
50:32products are being featured and is
50:34their presence of sales and discounts.
50:36So what we want to do is we you know we
50:38could use human coders to identify them,
50:40but we wanted to know can we use
50:43supervised machine learning to
50:44identify these key areas that could
50:47inform E cigarette prevention?
50:49So what is machine learning?
50:51Machine learning is powerful and it could
50:52be used to examine a large data set.
50:54So in this case large,
50:55many videos machine learning
50:57has been used to examine social
50:59media content around tobacco use.
51:01However,
51:02no studies have examined YouTube
51:04videos using machine learning.
51:06So this is a quick overview of
51:09what a machine learning does,
51:11so using an algorithm to it uses
51:15an algorithm to predict something.
51:16So in this case,
51:17if we're interested in it saying you know,
51:19can we use machine learning to to to
51:21identify if a video featuring an E
51:24cigarette first we need to teach the
51:26algorithm what an E cigarette is, right?
51:28So we we teach it, if it's jewel,
51:30if it's east, sick, if it's vape,
51:32then it's considered an E cigarette
51:34and this is A and this is,
51:35this data set is now.
51:37Used to train the machine learning
51:40algorithm and the algorithm learns
51:42from this example data set and later
51:45uses a different data set to predict
51:47whether they could identify an E cigarette.
51:50So if it correctly identify
51:51that there is an issue,
51:53regret that he's a successful model.
51:55If it fails to identify where
51:57if an E cigarette exists,
51:59when it doesn't,
52:00then we reach train this machine
52:02article rhythm until we could
52:04achieve a successful classification.
52:09So in our study, this is a model
52:11performance of our machine learning
52:13models for each of the four categories,
52:15F1 score is a measure of test accuracy.
52:18It's calculated from the
52:19precision and recall of a test.
52:23And this is a like a pretty good
52:25score considering the complexity of
52:27the themes that we were identifying.
52:31So what do we find?
52:32So this is a little more detailed
52:35look into video themes that we use
52:37in this case study versus our study.
52:40One that's what we have more themes here,
52:42and we also similarly identify the
52:44product views were the most common.
52:45And if you see a picture image here,
52:47this is an example of what a
52:49product review look like, right?
52:50This is Jewel starter Kit
52:52unboxing and review.
52:53And we also found that 72nd highest
52:56video theme was modified video that
52:59teaches people how to modify and
53:01informational videos on how to use
53:03health information was 11% other
53:06themes that were still ysaguirre.
53:099% of marijuana related things
53:11was 6% and other irrelevant theme
53:14which is like non E cigarette
53:16theme for five percent 5.6% and
53:18vape chicks was one point 1%.
53:23So product type, so this is so this
53:24is all the different types of products
53:27that we identified through machine
53:29learning and and what this actually
53:31shows is that there are a variety of
53:34different types of E cigarette products
53:36that are being featured on on YouTube.
53:39So who are the people who
53:41are uploading these videos?
53:4354% were weighed enthusiasm,
53:44so who are big enthusiasts?
53:46These are independent users who post
53:49almost exclusively about bathing.
53:51So when you go to the channel page to see
53:53what kind of videos they've uploaded,
53:54it was mostly related to vaping,
53:56but they were not directly
53:58connected to vaping company,
54:00so we cannot verify that
54:01their influences or not.
54:03So these are some examples of like
54:05account of people who've a person.
54:08Vape enthusiasts of channel page.
54:10As you could see,
54:11all the contents related to vaping.
54:13This is problematic because when
54:15it comes to regulating content,
54:17you cannot regulate private users, right?
54:21You can't tell the regular
54:22person to say you know.
54:23Don't post things about vaping.
54:25However,
54:25you could regulate influencers
54:27who get paid by the industry to
54:29post their products and the the.
54:31The difficulty with vape enthusiasts
54:33is that there's no way to tell
54:35who are vape enthusiast,
54:36who are influencers and her regular users.
54:4021% are stores,
54:4112% is other sources and six point 4% of
54:45medical community and 6% of private users.
54:52So 59% of video did not have any
54:54discount or links 34% of the videos
54:58had external links for purchasing
55:00and 5% or have other discount methods
55:03and one point 7% had discount.
55:04So this is a screenshot of of of
55:08instructional videos like beginning
55:10beginners vaping tip that also had
55:13a link that you could purchase
55:15as well as a coupon code.
55:17For purchasing,
55:18So what do we find in this study?
55:22We found that I complicated things
55:25relevant to E cigarettes could be
55:28identified using machine learning and
55:30fictitious youth viewer profiles on YouTube.
55:32We identified videos that violated
55:35YouTube tobacco policy restricting
55:37promotional content to underage minors,
55:39such as product reviews and purchasing links.
55:42Again, there was a high level
55:43of industry presence and such
55:45as faith enthusiast at stores.
55:49So overall conclusions, you know.
55:51Mixed methods such as qualitative
55:54analysis using human labellers and
55:56computational methods can really reveal
55:58E cigarette use content to inform youth,
56:01tobacco prevention and social media has
56:04really a really rich data and has a good.
56:09You know you could have a really good
56:12understanding of youth behaviors as well as
56:14promotion and sales that youth can access.
56:16And again, this is our current
56:18occurring a lot on YouTube as well
56:20as on other social media platforms.
56:22And to prevent youth E cigarette uptake,
56:25regulation of social media,
56:27a promotion that occurs in
56:29social media is really needed.
56:32So you know this is one example of
56:34how social media could be leveraged
56:36using qualitative and computation
56:38method to understand certain
56:39behaviors that could prevent KENS.
56:42Has cancer prevention
56:43implications like tobacco use?
56:45But certainly this this type of methods could
56:48be used to understand other behaviors that
56:51has direct implications to preventing cancer,
56:54such as, you know,
56:55physical activity,
56:56diet, obesity as well.
56:59So I'd like to acknowledge our funding
57:02stores as well as Yale Tobacco Center
57:04of the Study on Tobacco Regulation,
57:07tobacco product of Youth in addiction and
57:11also our team in University, Texas Austin,
57:14who is leading the computational methods.
57:19So thank you for your attention.
57:22Thanks Doctor Kang that
57:24that was that was great.
57:26And it's open for questions,
57:29please put him in the chat.
57:30I know we only have a few minutes,
57:31but maybe we could stay
57:33over for a minute or two.
57:35If people have questions.
57:39Have you? Reached out to YouTube and showed
57:45them your data and asked whether they,
57:47I mean it does sound like there's
57:50clear you have clear evidence that
57:52their policies are being violated.
57:54Presumably they have the computational
57:56firepower to be able to do similar things.
58:00Is it something that they may
58:02be convinced to look into?
58:04Yeah, that's a great question.
58:05You know, I have a paper cut
58:07currently under review that's
58:09looking at all of the self imposed.
58:11Social media policy.
58:12Across all the all the social media platforms
58:15on tobacco and and not surprisingly,
58:17you know all of the social media
58:19platforms that do have these policies.
58:22They're not being enforced,
58:23so so hopefully you know this will
58:25bring some more greater attention.
58:27Aside from You Tube.
58:28But just looking at all the social media
58:31platforms and what more could be done.
58:33Yeah, and I think could get,
58:35you know,
58:35I think 1 translational component
58:37is that we publish in peer review
58:39journals and a lot of this information
58:41don't get out into the bigger world
58:43and I think just doing some of
58:44that legwork might be important in
58:46getting some of these attention
58:47for two social media platforms.
58:51It's important work.
58:54So it's it's a few minutes
58:55after the hour doesn't look like
58:57there's any more questions.
58:58So again, thank you to both
59:00the the presenters for very
59:02interesting discussion and.
59:04We'll see you at the next grand rounds.
59:06Thank thank you.
59:09Thank you.