Jason J. Liu

This paper uses wearable data and AI to identify genetic associations with psychiatric disorders by creating intermediate phenotypes from continuous physiological features.

The paper processes single-nuclei multi-omics data from 388 human brains to identify regulatory elements and build cell-type networks, revealing changes in aging and neuropsychiatric disorders.

The paper identifies and validates enhancer regions in primary human neural progenitor cells using capture STARR-seq, linking them to genes involved in psychiatric disease pathways.

The paper introduces the EN-TEx resource, which provides personal epigenomes and variant-impact models across multiple tissues, revealing allele-specific loci and their associations with GWAS loci.

Adding 12 million AlphaFold2-predicted structures to training lifts native sequence recovery on held-out backbones to 51.6%, versus 42.2% for the best experimental-only model; likelihoods score mutation effects zero-shot.

This paper applies Bayesian structural time series to biomedical sensor data to evaluate interventions, showing its effectiveness in assessing treatment impacts with complex covariates.

A single pre-trained protein language model, ESM-1v, scores mutations zero-shot about as well as family-specific DeepSequence across 41 deep mutational scans, with no experimental labels or retraining.

DECODE improves enhancer prediction and boundary detection using deep learning and functional assays, achieving higher accuracy and reduced annotation size.

The paper introduces a Neural Potts Model that predicts pairwise coupling matrices for protein families using amortized optimization, showing improved performance on low-depth multiple sequence alignments.

The paper trains a deep language model on 250 million protein sequences to learn biological representations that capture structural and functional properties.

One protein language model trained across many families on whole multiple sequence alignments, with interleaved row and column attention, reportedly outperforms unsupervised structure-learning methods by a wide margin.

SCAN-ATAC-Sim simulates scATAC-seq data from bulk ATAC-seq to benchmark analysis methods with known cell-type labels.

The paper introduces STARRPeaker, a processing pipeline for STARR-seq data that improves enhancer detection by addressing sequencing biases and RNA stability effects.

NIMBus identifies mutational hotspots by integrating mutation data with genomic features to account for heterogeneity, revealing known drivers and regulatory disruptions in cancer.

RADAR identifies variants affecting RNA-binding protein function by integrating conservation, RNA structure, and tissue-specific data to prioritize noncoding variants.

The paper develops a custom ENCODE annotation for cancer genomics, highlighting TF-RBP interactions and network rewiring during oncogenesis, and validates findings using experimental assays.

DiNeR identifies significant changes in transcription factor co-regulation networks by analyzing ChIP-seq data from disease and normal samples, revealing key disease-associated regulators.

The paper develops a recurrent neural network to predict splicing outcomes using epigenetic signals, showing that spatiotemporal chromatin context is key to splicing regulation.

TopicNet measures transcriptional regulatory network change by abstracting gene regulatory topics and quantifying rewiring scores between cellular states, revealing TFs with significant network connectivity changes and their impact on cancer survival.

GRAM predicts the molecular effect of non-coding variants in a cell-type specific manner using SELEX and expression profiles, achieving high performance on MPRA and luciferase assays.