JJ Gao

Mount Sinai writes "Mount Sinai Researchers Use AI and Lab Tests to Predict Genetic Disease Risk - AI model uses common data to gauge risk from rare genetic variants" https://lnkd.in/eFrafiUF. #mountsinai #artificialintelligence #research #healthcare #geneticdiseaserisk #aimodels #labtests #aitests

1

We report new work from our group in collaboration with Google's genomics team. A challenge with genetic discovery for heart failure is that its classification within large scale biobanks is extremely vague. This has led to the surprising situation where, in 2026, there are only 2 validated GWAS loci for heart failure with preserved ejection fraction, a sub-group that accounts for HALF of the total heart failure patients, heart failure being the single most common admitting diagnosis in patients over the age of 65 (more than 30 million people globally). We need to do better. One approach is to use machine learning to develop a precision phenotype (for this you need rich health care system data from many patients, not biobank data). Then, you can take that probabilistic framework and apply it to biobanks which, while less enriched for patients, provide the power of large scale and deep genomics and proteomics. This was the idea that Jack W O'Sullivan MD, PhD and a large collaborating team sought to pursue in this work. The findings went beyond all our expectations. First, the team increased the number of loci confidently associated with HFpEF almost 50 fold! We report 99 loci, up from 2. Second, by integrating multi-omics in a causal inference framework, we prioritize gene-protein pathways as candidates for therapeutic development while de-prioritizing non-causal targets, reclassifying them as biomarkers. How could we tell we were on track? We noted first that three clinical trials for one of our deprioritized targets (MPO) were negative (genomics really can save money in drug discovery). We also validated one of our prioritized targets in a novel clinical study. In conclusion, we hope this work accelerates our understanding of heart failure. We believe this approach to precision phenotyping can extend the reach of our global biobanks in many directions. Finally, we believe incorporating causal inference in a multiomic framework can directly inform investment decisions in drug discovery pipelines. Congratulations to the entire team and special thanks to our Google collaborators who are the co-leaders of this work. Taedong (Ted) Yun Cory McLean Andrew Carroll Paper: https://lnkd.in/gamstJQM

532

17 Comments

In June, when we released Boltz-2—a new open-source AI model developed with Massachusetts Institute of Technology—I talked about its potential to accelerate drug discovery by predicting both structure and binding affinity. Now, our team is showing how we’re putting it to use. We’re excited to introduce a new paper that outlines Boltz-ABFE, a part of our Recursion OS 2.0 platform that addresses a major bottleneck in drug discovery. The challenge? The most accurate, physics-based methods for predicting how well a drug will bind to its target protein require a high-quality 3D crystal structure, which is often unavailable in the early stages of discovery. In our latest paper, we show how we combine Boltz-2's ability to generate accurate protein-ligand structures with a new, automated protocol. It’s an approach that makes gold-standard binding affinity predictions possible without needing an experimental crystal structure, allowing us to accelerate the identification and optimization of new drug candidates. Congratulations to our team on this important work! Stephan Thaler William (Zhiyi) Wu, @William G, @Richard T. Bradshaw, Gail Bartlett, Prudencio Tossou, and Geoffrey Wood Read the paper: https://lnkd.in/eCW-ZCyp PS - to hear more about Boltz-2, join Regina Barzilay and me this Thursday on TechBio Talks! You can find the episode on YouTube, Spotify, or wherever you get your podcasts. #DrugDiscovery #AI #FEP #TechBioTalks

223

𝗧𝗵𝗲 𝗲𝗿𝗮 𝗼𝗳 𝘀𝗶𝗹𝗼𝗲𝗱 𝗔𝗜 𝗺𝗼𝗱𝗲𝗹𝘀 𝗶𝗻 𝗰𝗮𝗻𝗰𝗲𝗿 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵 𝗶𝘀 𝗼𝘃𝗲𝗿. The transition from classical machine learning to 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗠𝗼𝗱𝗲𝗹𝘀 (𝗙𝗠𝘀) marks a fundamental shift in how we approach precision medicine. Integrating genomics, proteomics, imaging, and clinical data is no longer optional. It is the only way to unravel the complexity of cancer. I am thrilled to share that our latest research, "𝗙𝗿𝗼𝗺 𝗰𝗹𝗮𝘀𝘀𝗶𝗰𝗮𝗹 𝗺𝗮𝗰𝗵𝗶𝗻𝗲 𝗹𝗲𝗮𝗿𝗻𝗶𝗻𝗴 𝘁𝗼 𝗲𝗺𝗲𝗿𝗴𝗶𝗻𝗴 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹𝘀: 𝗿𝗲𝘃𝗶𝗲𝘄 𝗼𝗻 𝗺𝘂𝗹𝘁𝗶𝗺𝗼𝗱𝗮𝗹 𝗱𝗮𝘁𝗮 𝗶𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻 𝗳𝗼𝗿 𝗰𝗮𝗻𝗰𝗲𝗿 𝗿𝗲𝘀𝗲𝗮𝗿𝗰𝗵", has officially been published in 𝗔𝗿𝘁𝗶𝗳𝗶𝗰𝗶𝗮𝗹 𝗜𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲 𝗥𝗲𝘃𝗶𝗲𝘄. In this paper, we provide the first systematic map of the transition from conventional machine learning to advanced 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗠𝗼𝗱𝗲𝗹𝘀 (𝗙𝗠𝘀). We explore: 🚀 The shift from task-specific models to large-scale pre-trained backbones. 🧬 Holistic strategies for multi-omics and advanced imaging integration. 🗺️ A roadmap for the forthcoming era of large-scale AI models in oncology. 🎯 Our goal is to provide researchers and clinicians with the essential groundwork to revolutionize cancer diagnosis, prognosis, and biomarker discovery. 📖 𝗙𝘂𝗹𝗹-𝘁𝗲𝘅𝘁 𝗔𝗰𝗰𝗲𝘀𝘀: 🔗 https://rdcu.be/e73oj 💻 𝗘𝘅𝗽𝗹𝗼𝗿𝗲 𝗼𝘂𝗿 𝗚𝗶𝘁𝗛𝘂𝗯 𝗿𝗲𝗽𝗼𝘀𝗶𝘁𝗼𝗿𝘆: https://lnkd.in/eGNbBBN9 A huge thank you to my incredible co-authors: Muhammad Waqas Sheikh, PhD, Maliazurina B. Saad, Eman Showkatian, Rukhmini Roy, Hui Xu, Wentao Li, Joe Chang, Zhongxing Liao, Cara Haymaker, Luisa M. Solis Soto, @Carol C. Wu @Natalie I. Vokes, Xiuning Le, @Lauren A. Byers @Don Gibbons @John V. Heymach, Jianjun Zhang, Jia Wu #AI #MachineLearning #CancerResearch #FoundationModels #DeepLearning #PrecisionMedicine #MultimodalData #AIinMedicine #SpringerNature #Oncology

47

4 Comments

One of our core areas of work is curating datasets, and this also represents our use of LLMs for label extraction. When we begun labeling our Emory CXR dataset, we extracted both CheXpert labels and LLM labels. At that point we needed to make a decision - will we keep chasing every new LLM update (while remaining cost conscious) to update our labels or what was going to be our guide? We share our experience in this paper - https://rdcu.be/ekNCl led by Bardia Khosravi. We developed #Radprompter to help with tracking and versioning our prompts with new updates. We found - 1. LLM-based labeling outperformed the CheXpert labeler, with the best LLM achieving 95% sensitivity for fracture detection versus CheXpert’s 51%. 2. Larger models showed better sensitivity, while chain-of-thought (COT) prompting had variable effects. 3. Image classifiers showed resilience to labeling noise when tested externally. Simply stated - We can rely on using the smallest models with good performance. Be careful with COT as the models tend to "overthink" resulting in inaccuracies. For downstream task model training, LLM extracted labels are almost as good as human extracted labels - which means you can direct your resources to the test dataset. We found that the models trained on these labels were very robust to label noise.

73

3 Comments

🔬 What does a typical budget look like for a Phase 3 oncology clinical trial? Let’s unpack a detailed example. Designing and executing a large-scale oncology trial is a complex and costly endeavor. Here's a real-world example that models the budget for a randomized, two-arm, commercial Phase 3 trial evaluating Drug 1 + Drug 2 vs. Drug 2 alone in second-line advanced tumors across 25 sites in two EU countries. 👥 Patient enrollment: 350 patients 🗓️ Timeline: 66 months total Start-up: 6 months Recruitment: 36 months Per-patient treatment: 6 months Survival follow-up: 12 months (last patient in) Close-out: 6 months 🧩 Key CRO-driven activities and associated costs: 📌 Site Management & Monitoring Ongoing CRA support (email, calls, issue escalation): $950K Onsite monitoring (every 2 months/site): $1.07M Site initiation visits, activation, document access: $42K Close-out visits at all sites: $40K 📌 Project Oversight Centralized project management (scheduling, tracking, finance, sponsor updates): $1.02M Communication between central CRO & remote CRAs/Sponsors: $89K 📌 Regulatory & Start-Up Ethics & RA submissions, study amendments, insurance, progress reports: $64K Site identification & selection visits: $46K Contracting & site payment setup: $64K 📌 Medical, Data, and Document Management Medical writing (protocols, ICFs, CSR, publications): $25K Document & TMF management (CTAs, eTMF, ISF, reconciliation): $149K Data management (EDC setup, validation, coding, reconciliation): $435K Statistics (SAP, programming, sample size calc., SDTM/ADaM): $150K 📌 Safety, Logistics & QC Drug safety (SAE/SUSAR handling, safety reports): $30K Clinical & biological sample logistics (excluding pass-throughs): $152K Clinical supplies (lab kits, tracking, shipments): $22K Quality control (protocol deviations, inspections, eTMF reviews): $70K 📌 Pass-Through Costs (💸 largest line item – $8.5M) These include: Trial insurance (per country) Site payments per enrolled patient Lab kits, EDC licenses, shipping, EC/RA fees Drug distribution & depot fees Imaging platforms, DSMB fees, translations, etc. 💡 Total Budget: $12,929,155 👉 CRO operational costs: ~$4.4M 👉 Pass-throughs: ~$8.5M 📣 Why is this important? This example reflects the full-stack reality of global trial operations, blending regulatory compliance, patient-centric logistics, and data integrity, all while managing cross-border coordination and timelines. For sponsors, this means clarity on where their investment flows. 🔗 Whether you’re planning a trial or optimizing one in progress, understanding cost anatomy is crucial. 💬 I'd love to hear from peers in clinical ops, biotech, and CROs, does this budget reflect your experience? What’s rising fastest in your budgets today? #ClinicalTrials #OncologyResearch #ClinicalOperations #Biotech #Pharma #CRO #TrialBudget #ProjectManagement #DrugDevelopment #ClinicalResearch #MedicalWriting #ClinicalDataManagement #RegulatoryAffairs #HealthTech

8