Deep learning on
#graphs
is poised to address major gaps in biology and medicine
In Nature Biomed Eng
@natBME
, we describe the next generation of
#graphAI
#GNN
methods and opportunities that build models from data using structure, geometry & knowledge
Excited to share our
@Nature
paper on the role of AI in scientific discovery 🌟🔬
#AI4Science
AI is transforming discovery across sciences 🤖🔍 From hypothesis generation to data interpretation, it is reshaping all stages of research in ways we could not imagine using…
Can we infuse
#structure
into a time series (
#TS
) model from a diverse dataset so as to greatly improve
#generalization
on new TS coming from different datasets?
Yes, via a new principle called
#Representational
Time-Frequency Consistency (TF-C)
1/3
Survey on Representation Learning for Networks in Biology and Medicine
Long-standing principles of biomed nets (often unspoken in ML) provide grounding for representation learning, explain successes & limitations
@_michellemli
@KexinHuang5
#netbio
#GNN
#ML
Introducing PINNACLE, a contextual graph AI model for comprehensive protein understanding
PINNACLE dynamically adjusts its outputs based on molecular contexts in which it operates
Providing outputs tailored to molecular contexts is essential for broader use of foundational…
Introducing PDGrapher - Combinatorial prediction of therapeutically useful chemical and genetic perturbations using causally-inspired neural networks
Many methods learn responses to perturbations, but PDGrapher is addressing the inverse problem, which is to infer the perturbagen…
With
@payal_chandak
and
@KexinHuang5
, we are excited to share PrimeKG, a precision medicine-oriented knowledge graph providing holistic and multimodal view of human disease (1/8)
Excited to share preprint on Therapeutics Data Commons!
Paper:
Website:
TDC is a unifying framework across the entire range of
#therapeutics
#ML
. Ecosystem of tools, leaderboards & community resr
66 ML-ready datasets
22 ML tasks
How can we delete data from a big model without training the model from scratch and sacrificing performance?
Having models unlearn is notoriously difficult
Our new
#ICLR2023
paper introduces a general strategy for unlearning on graphs
1/ Have you ever wondered why AI models struggle to perform well when faced with new data outside their training set? This can be due to feature and label shifts which we can address using
#ICML2023
Raincoat
@icmlconf
@harvard
@harvard_data
@MITLL
@MIT
Excited to share the blueprint for multimodal graph learning in
@NatMachIntell
We envision graph AI playing a key role in future image-intensive, knowledge-grounded, and language-intensive systems 1/5
I was very happy to receive
@harvardmed
Young Mentor Award today.
Feeling fortunate to be working together with so many outstanding students
@HarvardDBMI
and am excited to see the bright future they are creating for themselves and our world
Excited to be elected to
@ELLISforEurope
as ELLIS Scholar in Geometric Deep Learning. Can't wait for new opportunities to further graph ML research on both sides of the Atlantic.
Many thanks to the ELLIS community for the support, esp
@mmbronstein
@maxwelling
@tacocohen
Excited to share our
@NeurIPSConf
papers accepted as spotlights!
@ZaixiZhang
@oq_35
#NeurIPS2023
Full-Atom Protein Pocket Design via Iterative Refinement
🧬 Designing functional proteins for specific ligand binding is key for therapy & bio-engineering. A���
🚨 JOB ALERT 🚨
We are building the next generation of Therapeutics Commons
@ProjectTDC
!
Our vision is to lay the foundations for AI & therapeutics, eventually enabling AI to learn on its own and acquire knowledge through continual refinement
We are seeking postdoctoral…
Today we share a paper introducing graph neural network
#GNN
powered mutual interactors to predict
#molecular
#phenotypes
Excited about this paper because it shows how thoughtful probing of AI model behavior can pave the way to improvements on many downstream tasks [1/6]
Our review+perspective paper on "Current and future directions in network biology" is online (). This has been a Herculean effort by many co-authors, thank you all! Special thanks to
@marinkazitnik
,
@_michellemli
,
@aydin_wells
, and all section coordinators.
We are thrilled to announce the National Symposium on Drugs for Future Pandemics (Nov 17-18)! Tune in for two days of visionary talks by stellar speakers. Registration is free and open
#futuretx20
Whoa, the current
@TheEconomist
edition reports on how AI can turbocharge scientific progress and lead to a golden age of discovery, echoing key insights from our recent
@Nature
paper,
Debate about AI often focuses on potential dangers: algorithmic bias, the mass destruction of jobs or even the extinction of humanity. But as some fret about dystopian scenarios, others are focusing on potential rewards
Live now: “Graph Representation Learning for Biomedical Discovery” by
@marinkazitnik
at
@logml2021
. Tune it to hear from an expert in the field how geometric deep learning is powering breakthroughs in biomedical discovery. Stream:
#MachineLearning
What could science at digital speed look like? 🚀
AI is poised to supercharge scientific discovery as we know it, by:
🔮 Exploring theories
🧪 Designing experiments
🔍 Analysing data
Find out how in
@Nature
. ⬇️
Mark your calendars for Nov 17-18! Join the discussion with leading experts in CS, bio, med, automation, and regulation about ways to innovate rapid therapeutics for future pandemics. Registration is free & open
#futuretx20
Submit your latest research to Graph Representation Learning Workshop at
#ICML2020
Deadline: May 29
We encourage submissions on
#graph
representation learning, geometric deep learning, interdisciplinary applications, benchmarks, and research aiming to mitigate
#COVID19
Excited about Graph Representation Learning and its application across the sciences? Our Research Scientists
@PetarV_93
and
@jhamrick
are co-organising an
#icml2020
workshop on the topic -- consider submitting your latest research by 29 May! More information below:
Great work,
@_camiloruiz
! We develop a multiscale interactome approach to explain disease treatments. It predicts drug-disease treatments, identifies proteins and biological functions related to treatment, and identifies genes that alter treatment's efficacy & adverse reactions.
Congrats to
@payal_chandak
and
@KexinHuang5
on publishing PrimeKG, multimodal knowledge graph for precision medicine in
@ScientificData
🔬 🎉
Stay ahead of the curve with PrimeKG as it's continually updated with the latest data
With
@payal_chandak
and
@KexinHuang5
, we are excited to share PrimeKG, a precision medicine-oriented knowledge graph providing holistic and multimodal view of human disease (1/8)
Dr.
@marinkazitnik
is an Assistant Professor of Biomedical Informatics
@harvardmed
with additional appointments
@harvard_data
and
@broadinstitute
of Harvard and MIT. Dr. Zitnik works on infusing knowledge, structure, and geometry into machine learning models.
We introduce and experiment with three new graph datasets comprising of high-stakes decision-making applications. Our results show that NIFTY improves the fairness and stability of SOTA GNNs by 92.01% and 60.87%, respectively, without sacrificing predictive performance [7/n]
It's an exciting double conference week,
#ISMBECCB2023
and
#ICML2023
. If you're attending, these talks and presentations are not to be missed!
At
@iscb
,
@_michellemli
will talk about her work with
@Emily_Alsentzer
on SHEPHERD, which is our latest zero-shot deep learning model…
Excited to share a new preprint on predicting molecular
#interactions
with skip-graph networks
Skip-
#GNN
uses 2nd-order interactions, which proved incredibly useful in
#bionets
over the last decade
Led by a fantastic student
@KexinHuang5
@harvard_data
Genes can participate in multiple independent biological functions, a foundational genetic principle known as pleiotropy. We show how sparse approximation & learning can decipher pleiotropy in high-dim gene perturb datasets. Great work,
@joshbiology
!
#MLCB2021
Ever felt your favorite gene has *more than one function* after perturbing it in many cell contexts? Today at
#MLCB2021
, I'll discuss automatic genotype-phenotype inference from high-dimensional gene perturbation data using sparse approximation: [a 🧵]
How do biological
#networks
change with
#evolution
? Our study just published in
@PNASNews
shows that evolution leads to resilient protein interactomes, which, in turn, are beneficial for organisms w/
@jure
, M.W. Feldman et al.
Here it is: the first Learning on Graphs Conference! 🎊
We think this new venue will be valuable for the Graph/Geometric Machine Learning community.
What makes it so important+unique? See our blog post!
1/6
Can we infuse
#structure
into a time series (
#TS
) model from a diverse dataset so as to greatly improve
#generalization
on new TS coming from different datasets?
Yes, via a new principle called
#Representational
Time-Frequency Consistency (TF-C)
1/3
Looking forward to meeting you all
@kdd_news
#KDD2022
- On Monday I will give keynotes on:
Infusing Structure and Knowledge into Biomedical AI (10:45am)
Graph-Guided Networks for Complex Time Series (14:20am)
🔍 Join Professor
@marinkazitnik
's lab as a Postdoc
#Research
Fellow in
#AI
for Cancer Drug Discovery at
@Harvard
! Apply to lead the design, development, and implementation of novel AI methods for the analysis of clinical and biomarker
#data
in oncology:
Congratulations to Ruth Ley for obtaining the Otto Bayer Award 2020! Congratulations to Julia Mahamid, Josep Cornella, Nikolai Franzmeier and Marinka Zitnik for obtaining the Early Excellence in Science Awards 2020!
@CornellaLab
@nfranzme
@marinkazitnik
Excited about our new work accepted to
#ICLR2020
as full paper with spotlight!
@iclr_conf
TL;DR: We develop strategies for pre-training Graph Neural Networks and study their effectiveness on multiple datasets, GNN architectures, and diverse downstream tasks
#molecules
#proteins
Subgraph Neural Networks
GNNGuard: Defending GNNs against Adversarial Attacks
Graph Meta Learning via Local Subgraphs
Open Graph Benchmark: Datasets for ML on Graphs
Thank you
@BiswasFamilyFdn
for your vision and generosity
Learn more about our project CURE-Bench to build and evaluate all-disease foundation models for identifying clinically relevant drug repurposing hits
Thankful to
@BiswasFamilyFdn
for supporting…
Excited to see this published: We examined 10,443,476
#adverse
#drug
event reports, spanning 19,193 adverse events and 3,624 drugs, to extract insights into safe
#medication
use & how adverse events vary across
#patient
groups
Detailed tweetorial to follow shortly
#data
#AI
In a recent Article,
@xiangzhang1015
,
@marissa_sumathi
and
@marinkazitnik
analyze large-scale patient safety data to reveal demographic disparities of drug safety and identify at-risk patients during a pandemic.
Coming soon: NEJM AI, a new journal from NEJM Group.
NEJM AI aims to identify and evaluate state-of-the-art applications of artificial intelligence to clinical medicine. Learn more about the new journal:
We are thrilled to release the Open Graph Benchmark!
OGB contains numerous biomedical datasets, including protein interaction nets, cross-species graphs, drug-drug interaction nets, and biomedical knowledge graphs
#ML
#graphs
#networks
Super excited to share Open Graph Benchmark (OGB)! OGB provides large-scale, diverse graph datasets to catalyze graph ML research. The datasets are easily accessible via OGB Python package with unified evaluation protocols and public leaderboards.
Paper:
AI and science:
@Richvn
@Nature
polled 1,600 researchers around the world about their views on the rise of AI in science, including machine learning and generative AI
#AI4Science
@AI_for_Science
Among those who used AI in their research, more than 25% felt that AI tools would…
Excited to present a tutorial on
#ML
for drug development and discovery at
@IJCAIconf
.
Jan 6, 7-10pm EST / Jan 7, 9am-12pm JST w/
@jimeng
and Danica Xiao
@IQVIA_global
So glad to see this! We started to tackle this challenge through Therapeutics Data Commons, an ecosystem of AI/ML-ready datasets for therapeutics, together with tools, leaderboards and community resources like meaningful data splits
The human genome is gradually unravelling its secrets 🎁
AlphaMissense model
@ScienceMagazine
: one more path lit up by deep learning in exploring the code of life 🧬
We now know with high confidence if 89% of ALL missense variants are benign or pathogenic
Key contributions🧵🧵
Phenomenal resource with ready-to-use embeddings and models through
@cziscience
Cell Census 🚀
Congrats to
@JCoolScience
@Paedugar
and the
#CELLxGENE
team 👏
Thanks for showcasing PINNACLE, our contextual AI model for single-cell protein biology, and scCIPHER, our multimodal AI…
🚨1/ New to CZ
#CELLxGENE
: models & embeddings that integrate up to 36M cells in the Census corpus.
Use embeddings to explore the corpus directly, or download the models to run your own data through them to enable direct comparisons to the reference. 🧵
Hello Twitter! This is the official Twitter account for the Workshop on Graph Learning Benchmarks (
#GLB
). We are pleased to announce that the 2nd GLB workshop will be held with the
#WebConf
2022. The CfP is out! The submission due is Feb 28, 2022.
More at
AI is catalyzing scientific discovery into health and disease.
To accelerate progress, we’re building one of the world’s largest computing systems dedicated to non-profit life science research that we’ll leverage to create predictive models of whole cells
Introducing AI for Science — a
@NeurIPSConf
2021 Workshop! Our workshop focuses on bridging the gaps between machine learning and science. We have a stellar lineup of speakers! Submit your work and sign up for mentorship program now:
Grateful for the chance to discuss AI's 2024 prospects with
@NatMachIntell
. We covered LLM progress, multimodal AI, multi-task agents, and the crucial issue of bridging the digital divide across communities and world regions
Appreciating the insightful exchange,
@LCVenema
!
Nature Machine Intelligence has turned 5! Many thanks to all colleagues, authors and referees for helping us shape the journal. Read our anniversary edition of AI Reflections - interviews with recent Comment and Perspective authors
To hear highlights about our ongoing research you can find us at these 4
@icmlconf
workshops
- Socially Responsible ML
- Theoretic Foundation, Criticism, Trends in Explainable AI
- Interpretable ML in Healthcare
- Computational Biology
#icml2021
Schedule:
Congratulations to the 101 recipients of the 2020
#AmazonResearchAwards
, who represent 59 universities in 13 countries. Each award supports the work of one to two graduate or postdoctoral students for one year, under the supervision of a faculty member.
Join us
@NeurIPSConf
2021 "AI for Science" Workshop — Monday, Dec 13, 8a-6p ET
@AI_for_Science
Great day to celebrate AI achievements in scientific discovery and highlight open challenges that need to be addressed to move the field forward
#AI4Science
📢📢1/ Thrilled to share
@ProjectTDC
Therapeutics Commons 0.4.0
We have a new interface allowing users to easily access and leverage large pre-trained models for direct prediction or fine-tuning on downstream tasks
Check out our
@huggingface
Hub
Excited to share how combining biological LLMs with knowledge graph AI enables zero-shot prediction in drug discovery 🎯💊
Thanks
@HarvardCMSA
for organizing this conference
We are very excited to be co-organizing a workshop at
#AAAI2021
on Trustworthy
#AI
for
#Healthcare
! We have a stellar lineup of speakers. Details to follow soon!
@RealAAAI
w/
@cmuptx
, Byron Wallace, Jennifer G. Dy, and Eric P Xing.
How well do your AI models perform on new molecular sequences?
@YEktefaie
🧵 👇
Understanding generalizability - how well an AI model works on new data - is crucial in biology. This challenge grows with foundation models, large pre-trained models that promise to better predict…
It is no secret there exists a generalizability problem in AI for biology. Despite all the advances in ML methods, the way we evaluate generalizability in ML models has not changed. How well does your ML model generalize across the entire spectrum of possible dataset splits?…
One big question in
#AI4Science
@axios
: Can we build more than a scoop machine?
Instead of having AI predict discoveries and who will make them, the researchers tuned the model to "avoid the crowd," and it found scientific blindspots — surprising combinations and predictions…
Congrats to graduate student Michelle Li
@_michellemli
@HarvardDBMI
on being awarded
@NSFGRFP
Graduate Fellowship! Well deserved! Thrilled to see what awesome things you do next.
Excited to share our new paper in Nature Chemical Biology
@nchembio
AI is poised to transform
#therapeutic
#science
The Commons is an initiative to access and evaluate
#AI
capability across therapeutic modalities and stages of discovery 1/4
Drug combinations can lead to new treatments with better efficacy🎯 and reduced toxicity. But, combinatorial screening is infeasible to scale up.
Can ML models solve this problem? Happy to share the 2nd Leaderboard: combination response prediction!
Info:
Therapeutics Data Commons is a collection of
#ML
tasks spread across areas of
#therapeutics
22 tasks
67 datasets
Therapeutics ML offers incredible opportunities for innovation and impact
TDC informs ML model dev, valid and transition into production & clinical implem. Join us!
@xiangzhang1015
@MITLL
@Harvard
@HarvardDBMI
Results suggest models can be trained on a source TS dataset and deployed on a range of target TS datasets without retraining, resulting in performance that is better than that of state-of-the-art & state-of-practice models 2/3
Congrats
@WangQianwenToo
for winning Best Paper Honorable Mention award
@ieeevis
🏆
Fusing
#visualization
with
#ML
is increasingly important to ensure
#AI
outputs are actionable, support transition into biomedical implementation, and more
Thankful for collab w
@ngehlenborg
📢 1/ We are delighted to have Marinka Zitnik
@marinkazitnik
from
@Harvard
with us for the next TrustML seminar on Thursday, July 22nd, 12pm ET! 🎉🎉
Register here: .
Check this thread for the speaker and talk details! 👇