←
Harvard Data Science Initiative3,805 posts
Harvard Data Science Initiative avatar

Harvard Data Science Initiative

@harvard_data

Your friendly neighborhood Data Science Initiative.

◷ Joined August 2017▧ 600 Photos and videos
Marinka Zitnik avatar
⟳ Harvard Data Science Initiative retweeted
Marinka Zitnik@marinkazitnik·22 Jul 2024

Excited to share our new paper on Contextual AI models for context-specific prediction in biology in @NatureMethods led by stellar @_michellemli rdcu.be/dOxQ7 Understanding how proteins work and developing new therapies requires knowing which cell types proteins act in and how they interact with each other. Predicting and optimizing protein targets must happen in context. We can draw a parallel with the polysemic word “apple,” whose meaning is resolved via the context of surrounding words. Just as one can “grow an apple” or “buy an apple,” the roles of genes and proteins are resolved via cell context—particularly the environment where a drug will operate. In this @naturemethods paper, we introduce PINNACLE, an approach that uses geometric deep learning to create context-aware protein models. Using a protein interaction dataset and a multi-organ single-cell atlas @cziscience, PINNACLE analyzes protein interactions in 156 cell types across 24 tissues, generating nearly 395,000 protein representations. PINNACLE dynamically adjusts its outputs to biological contexts. Excited to soon share with you models across @cziscience #CellxGene Discover datasets. It paves the way for a type of AI that can learn contexts to understand a given protein plus its surrounding environment and identify its many potential roles. Providing outputs tailored to biological contexts is essential for the broad use of foundation models in biology. We tested PINNACLE on tasks such as enhancing 3D structural representations of therapeutically relevant interactions in immunology, studying the effects of drugs across cell-type contexts, nominating therapeutic targets in a cell-type-specific manner, and zero-shot retrieval of tissue hierarchy. @harvard @HarvardDBMI @harvard_data @KempnerInst @Roche @MassGenBrigham @BrighamWomens Fantastic team of collaborators: @_michellemli, M Sumathipala, MQ Liang, A Valdeolivas, AN Ananthakrishnan @kat_liao @danmarbach Thanks to @DrArunimaSingh for editorial guidance Paper: nature.com/articles/s41592-0… Research Briefing: nature.com/articles/s41592-0… Code: github.com/mims-harvard/PINN… HF Space: huggingface.co/spaces/michel…

Post attachment
Soroush Saghafian avatar
⟳ Harvard Data Science Initiative retweeted
Soroush Saghafian@Soroush_Saghaf·17 Jul 2024

𝓠𝓾𝓮𝓼𝓽𝓲𝓸𝓷: How can #AI revolutionize #healthcare? Why it has not been as impactful? See some answers below and learn more about our work @PIASLab 👇👇👇

Artificial intelligence has the potential to revolutionize health care delivery, but several obstacles stand in the way to wider adoption of A.I. tools. Professor @Soroush_Saghaf, founder of @PIASLab, is working on solutions that would improve the design of these tools.

3:27
Harvard Kennedy School avatar
⟳ Harvard Data Science Initiative retweeted
Harvard Kennedy School@Kennedy_School·15 Jul 2024

Can a bad debate performance shift voter preference? New research by HKS's Matthew Baum and colleagues evaluates the impact of the first Trump-Biden presidential debate ken.sc/3LpoLCI

Berkman Klein Center for Internet & Society avatar
⟳ Harvard Data Science Initiative retweeted
Berkman Klein Center for Internet & Society@BKCHarvard·11 Jul 2024

ICYMI, the full exposition of this talk by @zittrain on renegade #AI agents is now available as an op-ed in @TheAtlantic: theatlantic.com/technology/a…

ControlAI@ControlAI3 Jul 2024

Legal scholar Jonathan Zittrain on the the risks of AI agents: "Give it a few goals, let it have a bank account, let it draw from the bank account to spend money on stuff ... and then who knows where it ends up?"

0:49
Harvard Statistics Department avatar
⟳ Harvard Data Science Initiative retweeted
Harvard Statistics Department@HarvardStats·11 Jul 2024

Imai & Li's paper (forthcoming in the Journal of Causal Inference) demonstrates that Neyman's methodology can be used to experimentally evaluate the efficacy of individualized treatment rules (ITRs), which are derived by modern causal ML algorithms: arxiv.org/abs/2404.17019

NEJM AI avatar
⟳ Harvard Data Science Initiative retweeted
NEJM AI@NEJM_AI·11 Jul 2024

Introducing a new artificial intelligence-enabled Perspective series that summarizes and extracts insights from the 𝘕𝘌𝘑𝘔 𝘈𝘐 Grand Rounds podcast episodes. Read the full editorial by Drs. @arjunmanrai and @AndrewLBeam: nejm.ai/45HhFmG

Post attachment