Debora Marks
@deboramarks
Followers
4K
Following
2K
Media
55
Statuses
1K
Mission: design of biological interventions for environment and human health (to save the world)
boston
Joined September 2008
Our new paper uses WaveNet-type method for “Protein Design and Variant Prediction” no alignments needed ! trains on only natural sequences with a generative autoregressive model on a dilated-CNN - collab with McMahon, Kruse and @AashishManglik 1/7
4
112
394
Do we need new ways to analyze biological sequences? Will it make a difference? Yes - .
2
60
211
Following the 200 million structures from #AlphaFold today - the Marks lab *may* release 200 BILLION structures of sequences only one mutation away from natural proteins.
4
9
203
Did you know you can publish in Nature before you even submit the paper?
9
9
164
Very proud LMRL was accepted for a second year at NeurIPS Can ML accelerate biomedical discovery? From medicines for people, to protein designs that eat plastics, to forecasting viral evolution, to human genetic variation. Come!.
2
35
133
Great to be awarded a NIH Director's Transformative Research Award with amazing team @chang_c_liu and Andrew Kruse .
9
8
127
How can we use #GenerativeAI for #ProteinDesign? - check out work from our labs @deboramarks and @yaringal .
0
26
124
Machine learning for synthetic immunology . and new antimicrobial discovery : hiring !
0
23
117
@ATinyGreenCell I suppose my lab - ha ha .Might also look back at our protein structure from sequence too - umm 2011?.
2
13
117
NEW #SARS_CoV_2 resource from Marks & Sander labs!. What effects will mutations have?.How are SARS-CoV-2 proteins folded? What interactions are important?. We're using ML to help accelerate #COVID research/therapies. Our results so far are LIVE at
4
50
104
Woohoo - Postdoc positions in the new Schmidt Center @ the Broad - in ML and biomedicine - choose any lab; come work in mine!.
3
27
101
Will this turn the gene regulation world around ? Great work by Barkai lab
4
38
96
Using deep mutational scanning to benchmark variant effect predictors and identify disease mutations | Molecular Systems Biology
1
22
91
Predicting effect of mutations using unsupervised generative models better than supervised. Worth repeating lesson in other bio areas
1
16
75
Can generative models of viral evolution to predict viral mutants that escape from antibodies ? - see poster/ spotlight talk at ICML WCB
0
17
73
Want to fold proteins? Predict effects of mutations - in silico deep mutational scans ? .
2
21
65
New paper: 'Optimal Design of Stochastic DNA Synthesis Protocols based on Generative Sequence Models' led by @EliWeinstein6
2
13
70
We want to build generative models to predict functional sequences: E.g. What future viral sequences will emerge? Check our paper at ICML 2021 led by grad student @EliWeinstein6 1/7
1
20
67
New paper yesterday by Marks Lab -Adam and John -@naturemethods - exploiting billions of evolutionary experiments to learn effects of mutations and sequence generation
1
20
62
Join us @ NeurIPS 2021 - workshop - LMRL Predict and Design Biology - structure, drugs, viral evolution, biological materials, climate change ? Submit paper/poster to Bring compute and biotech together to save the world :)
1
18
63
At NeurIPS? Come to our very first workshop - packed with exciting talks from ML and biology aficionados .
0
13
59
is recruiting an Associate Lab Director! An outstanding opportunity for an ambitious computational biologist to manage projects at the cutting edge of machine learning and genomics. Send CV to deboramarkslab@gmail.com.
2
21
58
Apply! @deboramarks lab is hiring together with @Jura_Bio - fab opportunity for ML postdoc to work on autoimmune disease with never-before-seen data. Apply at
1
34
61
@ylecun @ATinyGreenCell Although you don't cite our critical paper (same time) where we showed you could fold into correct 3D structural topology from the inferred contacts. This is the paper showed the power and gave a nice strong premise for developing AF -
1
13
59
Help! Trying to catch up on advances in small molecule binding to protein prediction methods - suggestions welcome !.
12
10
60
BEAR - a new generative model for whole genome sequences -scalable, theoretical guarantees; led by @EliWeinstein6 and Alan Amin #NeurIPS2021
0
15
56
Gorgeous paper Nathan Rollins Kelly Brock - 3D structure from deep mutagenesis scans
0
30
54
Shocking even to me - 88% AUCs from these microbiome papers can’t be trusted - data leakage ++ - time to check other ML- Bio claims
4
4
51
Fabulous - and tested - more than wishful thinking. Now - what shall we design ?.
This twirling work of science is special ✨. AFAIK, it's the first crystal structure of a functional #protein fully designed by #AI. A milestone in our quest to use language models to generate proteins that are unseen in nature & can function well in the real-world. Read below👇
1
3
52
Sweltering Sunday refreshed by a great community gathering ‘Bits in Bio’ in Boston - join us for future events!
1
5
52
Introducing Tranception - reaching new parts of the fitness universe.
Very pleased to announce that our paper “Tranception: protein fitness prediction with autoregressive transformers and inference-time retrieval” was accepted at ICML!.Paper: Repo: (1/10)
0
9
49
Facebook Proteins paper - 3D contact prediction with "only" 43% sequence identity. Oh wait, didn't folks do that 20, 30 years ago ? I am confused
2
15
49
Want to join us to design and generate molecules for sustainability? New probabilistic methods for biased learning?.
2
8
41
Let’s design enzymes and molecules to help save the world - I’m not being facetious.
1
3
45
Shout out to Liz Wood and Ray Jones who did a simply amazing job @lmrl_bio @NeurIPSConf today. And all the speakers and attendees.
2
3
43
Looking forward to reading Sinai and Kelsic “A primer on model-guided exploration of fitness landscapes for biological sequence design”
1
2
40
1
7
37
We’d love feedback . .
1/n I'm excited to share our preprint *Large-scale clinical interpretation of genetic variants using evolutionary data and deep learning* from a great collab with @NotinPascal @MafaldaFigDias @AidanNGomez @kpgbrock @yaringal (@OATML_Oxford) @deboramarks 🧵
1
8
29
Recruiting lab director/manager - Chance for a super organized comp literate person to guide a crazy friendly lab ! DM me.
0
15
28
Register and send abstract !.
0
0
28
Representation learning meets biology - the ultimate challenge.
0
1
29
Send abstract by Oct13th for @ NeurIPS 2021 - workshop - LMRL Predict and Design Biology - - structure, drugs, viral evolution, biological materials, climate change ?.
0
8
27
How well . or not. can we predict viral immune escape ? - new paper - see thread below- comments welcome.
Can one anticipate viral evolution enough to support timely intervention? Check out our new preprint on forecasting antibody escape.With @sarahgurev @NotinPascal @nooryoussef03 @_nathanrollins @sandercbio @yaringal (@OATML_Oxford), @deboramarks, 🧵1/10
0
6
25
Great work by members of my group - nanobody design too.
Super nice work from labmates Adam, June, Aaron and @deboramarks on using autoregressive models for protein design
0
10
23
Missing: My ML method is newer than yours, with a better name.
1
1
25
Yes yes - we make models that turn sequence data into actionable biotech - but bottleneck is being able to scale up in parallel - such an amazing opportunity . if community can come@together to do this - human disease, protein engineering … ….
But data alone is not value. To efficiently learn from massively parallelized data, we need access to massively parallelized compute. Hooray for @googlecloud, @awscloud, and @Azure! But we also need production software that abstracts scientific intent from low-level execution.
1
1
25
@RolandDunbrack Our recent paper on antibody library design was rejected by a reviewer because ‘it’s not like alpha fold’ - crazy times.
3
1
23
Predicting structure of designed protein is not harder than natural ones esp when designed have encoded the rules we know - take care with logic there folks . .
0
3
23
New paper - deep learning for sequences gives way to predict mutations and travel in seq space
1
19
23
EVE code now public - ‘Large-scale clinical interpretation of genetic variants using evolutionary data and deep learning’.
Our code for making EVE models is now public! .Code: Preprint: Website (update coming soon!): Recent MIA talk with @MafaldaFigDias: Tweetorial:
0
2
23
Oh, did I say?. ideal candidate is fluent in 50 languages, can fly unassisted and has a wicked sense of humour . .
1
0
20
Submit your cool new work in bio/ML to at loads of ops for students.
Submissions for #LMRL are live! Please share widely
1
3
22
An ultra-high affinity synthetic nanobody blocks SARS-CoV-2 infection by locking Spike into an inactive conformation .Nice work Michael, Peter, Aashish and all!.
0
4
20
@MoAlQuraishi Yep - We have a problem. Reification of method over the results seems to be endemic with ML in biology - let’s learn to have different measures of success. Ones that matter.
1
1
18
Awesome work by Eli Weinstein - Put in the model of your choosing. .
1
7
18
Join us! Open community - task driven :).
“Scientists launch 'Herculean' project creating atlas of human genome variants”. #ThrowbackThursday.
0
1
14
Yep - agree. And . looking ahead . I’ll reluctantly admit (some) biology maybe good prep for ML/AI? not only for bio-domain applications.
Which is why the best preparation for a PhD in ML/AI is not necessarily CS but electrical engineering, physics, or applied mathematics.
1
0
17
I love integrins: dynamic, information transmitting proteins; a well-timed award as we move from scRNA to cell-cell communication and physiology - the next 10yrs biotech.
Congratulations Richard O. Hynes, Erkki Ruoslahti, and Timothy A. Springer, 2022 #LaskerAward winners! — “for discoveries concerning the #integrins—key mediators of cell-matrix and cell-cell adhesion in physiology and disease”. @MIT @sbpdiscovery @harvardmed.#Lasker2022
0
3
15
2/2 Is data split for training fair? Where is comparison to state-of-art predictions for larger benchmark of mutation scan datasets? ( that are unsupervised from sequences alone) eg.
1
8
15
Totally has to become a thing - some great trail blazers out there like Steve and Lani at UCSF, now @MafaldaFigDias and @Jonnygfrazer starting together @CRGenomica . .
1.What if we explored different models for how labs could run? For example? the multi-PI lab. Presently every PI is like a decathlete: paper writing, grant writing, project management, budgeting, mentoring, etc. It would be so fun to run lab w/ other PIs & divide these chores up!.
1
1
16
Don’t test simple baseline models that run on a laptop against fancy ones on a TPU in case the simple ones work (sarcasm).
What is your most controversial computational biology opinion?. Most software should not produce any logging or status output to the console by default.
0
0
13
Beautiful work . .
Very pleased to share a new pre-print with @jmhernandez233 and @yaringal @OATML_Oxford.“Improving black-box optimization in VAE latent space using decoder uncertainty”.(1/)
0
2
12
And . biology.
Since @ylecun introduced his cake analogy at NeurIPS 5 years ago, unsupervised learning and the self-supervised approach in particular, has been a major focus of @facebookai and other AI labs. Progress has been steady and remarkable almost everywhere - NLU, MT, Speech, CV, Video
0
2
15
Primer on Medical and Population Genetics | Broad Institute - Nikki Thadani and Sarah Gurev from my lab - suggesting a way to predict immune escape early .
0
4
15
Thx to @nlarusstone for getting community going, to @elizahudson and @MorseCell for organizing the event and speakers-audience for a great back and forth.
0
0
13
Great to see EVE being real world useful
my 🤯 moment today seeing ML/big data have actual clinical impact: for ICU patient, re-analyzed 2010 genetic test w/ to ID pathogenic variant that may *finally* unify the symptoms. 12 years ago was called "likely not pathogenic". Progress is real.
0
1
13
@rfsquared @Dereklowe @DeepMind Deep mind professionalized homology modeling and prediction from sequences in an MSA. Really well. An academic group could not have got funding for this. Obviously. The overclaim is depressing and not needed.
1
0
13
Use to evaluate and design vaccines?.
Previous viral evolution can shed light on future immune evasion. EVEscape, now out in @Nature, is our first step in applying pre-pandemic data to forecast antibody escape through the course of a pandemic. 🧵1/6.
0
0
13
That’s why we have to find new ways to infer homology or similarity - and new ways to find structural plasticity.
@jinboxu_chicago @sokrypton @labriataphd @deboramarks @sandercbio This is also true for a lot of proteins outside CASP. In fact for many of the interesting proteins we work on. One may be fooled by the fact that the PDB is biased towards proteins with more homologues than a typical protein as @wim_vranken showed
1
3
10
@jmschreiber91 Worse than that - many methods cannot discover anything because ‘benchmarks’ are just inadequate for the goal.
2
2
11
Facebook policies antithetical to #CZI missions -lucky to be a grantee and proud to sign this letter
0
3
11
JOB: work with the Marks lab to find out cool things from sequences. Statistics, data handling and large doses of common sense needed.
0
14
11
Probably not. Does Canada ? Should I move ?.
Dear International Students in the US, Does the US really deserve you?.
0
0
11
Try repetition in biology data - it’s the new frontier.
🤣🤣.
0
0
11