Skip to content

S1 — Precision Medicine

Full session title: Precision Medicine: Integrating Large-Scale Data and Intermediate Phenotypes for Understanding Health and Treating Disease Papers: 10 (pp. 600–751) Session introduction: Brenner S.E. et al., pp. 596–599 · PDF · PMID 41758171

Organizers:

  • Steven E. Brenner — UC Berkeley
  • Nilah M. Ioannidis — UC Berkeley; UC Santa Cruz
  • Tayo Obafemi-Ajayi — Missouri State University
  • Anne O'Donnell-Luria — Boston Children's Hospital; Harvard Medical School; Broad Institute

Invited talk: Anne O'Donnell-Luria (Sun Jan 4, 9:10 a.m.) — title not printed in the schedule.

Reading this session

The organizing idea is the intermediate phenotype: something measured between genotype and diagnosis — a polygenic score, an imaging-derived endotype, an ECG waveform, a proteomic panel — used to make risk prediction work where genotype alone does not.

Polygenic scores, four ways. German et al. show polygenic risk improves generative forecasting of disease trajectories, not just point risk. Venkatesh et al. combine imaging-derived clinical endotypes with plasma proteomics and an external PRS for coronary microvascular disease. Cardone et al. add clinical, lifestyle, and social risk factors to polygenic scores for heart failure. Seagle et al. compare PRS-CSx against pruning-and-thresholding for polygenic partitioning of treatment-resistant hypertension. Read together, this is a session arguing that PRS is useful mainly in combination — with imaging, proteomics, or social data — rather than alone.

Interpretable deep learning. Rifat et al.'s BioLM-NET combines prior biological knowledge with contextual LLM gene embeddings over multi-omics data. Sethi et al. use prototype learning to build refined interpretable digital phenotypes from ECGs. Sriram et al.'s DeepDiff-SHAP does subgroup-specific causal hypothesis generation with conditional SHAP.

The pathology paper. Tamura et al. is the one to read if you care about digital pathology: weakly supervised neuron selection in sparse autoencoders over CLIP-derived pathology foundation-model embeddings, evaluated on Camelyon16 (breast-cancer lymph-node metastasis) and PANDA (prostate cancer grading). It is mechanistic-interpretability tooling applied to a pathology FM, and it explicitly demonstrates tumor-patch identification.

Papers

Title First author Pages PDF PMID
Integrating Polygenic Risk Improves Generative Forecasting of Disease Trajectories Chris German 600–611 german.pdf 41758172
Integrating Imaging-Derived Clinical Endotypes with Plasma Proteomics and External PRS Enhances Coronary Microvascular Disease Risk Prediction Rasika Venkatesh 612–628 venkatesh.pdf 41758173
Integrating Polygenic Scores with Clinical, Lifestyle, and Social Risk Factors to Improve Heart Failure Risk Prediction ⚠️ Katie Cardone 629–643 cardone.pdf 41758174
BioLM-NET: An Interpretable Deep Learning Model Combining Prior Biological Knowledge and Contextual LLM Gene Embeddings on Multi-Omics Data to Predict Disease Jubair Ibn Malik Rifat 644–663 rifat.pdf 41758175
Impact of Using PRS-CSx and Pruning and Thresholding for Polygenic Partitioning of Apparent Treatment Resistant Hypertension Hannah M. Seagle 664–678 seagle.pdf 41758176
Prototype Learning to Create Refined Interpretable Digital Phenotypes from ECGs Sahil Sethi 679–693 sethi.pdf 41758177
Deep Learning-based Classification of Patients with Postural Orthostatic Tachycardia Syndrome Using Wearable ECG and Accelerometer Data Hyunjun Choi 694–707 choi.pdf 41758178
Patch-Level Phenotype Identification Via Weakly Supervised Neuron Selection In Sparse Autoencoders for CLIP-Derived Pathology Embeddings Keita Tamura 708–721 tamura.pdf 41758179
DeepDiff-SHAP: Interpretable Deep Learning for Subgroup-Specific Causal Hypothesis Generation Using Conditional SHAP Aditya Sriram 722–737 sriram.pdf 41758180
Literature-Driven Extraction and Computational Prediction of Causal Statements Linking Genetic Variants to Biological Processes, Pathways and Phenotypes Jici Jiang 738–751 jiang.pdf 41758181

⚠️ cardone.pdf is not hyperlinked from the PSB proceedings TOC although the file exists and resolves. A link-following scrape will miss it silently.

Connections to the corpus

Sources