Jacob Matthew Schreiber

tangermeme provides tools to interpret deep learning models for cis-regulatory analysis, extracting human-readable insights from genomic predictions.

The paper designs tissue-specific enhancers for mouse embryos using deep learning models trained on chromatin accessibility data and validated enhancers.

This study quantifies intrinsic dCas9-DNA recognition using TANGO, identifying key factors affecting gRNA performance and revealing new aspects of dCas9 specificity.

This paper introduces PISA, a tool for interpreting sequence-to-function neural networks, enabling the discovery of nucleosome-positioning motifs and understanding cis-regulatory code through MNase-seq data.

The paper re-implements Tomtom as a faster version for motif similarity scoring, achieving over a thousand-fold speedup.

The paper predicts missing chromatin conformation contact maps across cell types and assays, exploring how 3D genome architecture varies.

The paper introduces Ledidi, a computational method for designing cis-regulatory elements by optimizing continuous sequences guided by deep learning models, validated experimentally for controlling transcriptional outcomes.

The paper introduces Icebear, a neural network framework for cross-species imputation and comparison of single-cell transcriptomic profiles, revealing evolutionary adaptations of X-chromosome upregulation in mammals.

The paper trains a deep learning model to predict transcription initiation from DNA sequences, revealing shared cis-regulatory logic and epistatic interactions across promoters and enhancers.

The paper uses deep learning to analyze chromatin accessibility and gene expression during reprogramming, revealing how TF concentrations and motif syntax influence cis-regulatory networks.

The paper develops L-HiC-Reg to predict long-range regulatory interactions in new cell lines and applies it to interpret GWAS SNPs in complex diseases.

The paper evaluates 23 imputation models for cross-cell type epigenomic data, finding quality assessment challenging due to distributional shifts, data availability, and redundant metrics.

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.

This paper introduces Yuzu, a method that accelerates in-silico saturation mutagenesis by using compressed sensing, reducing computation time for convolution-based models.

fastISM accelerates in-silico saturation mutagenesis for convolutional neural networks by 10x, making it feasible for large-scale sequence analysis.

This paper identifies a bias in machine learning models that predict genomic features across cell types when trained on data from the same genomic loci across multiple cell types.

The paper demonstrates using submodular optimization to create efficient, high-quality sketches of single-cell RNA-seq data.

The paper extends an imputation method to enable cross-species predictions by jointly modeling human and mouse epigenomic data, improving imputation accuracy for mouse and enabling zero-shot predictions for human assays in mouse.

The paper develops a method to prioritize genomic and epigenomic experiments using imputed data to maximize information gain, and introduces a software framework called Kiwano.

This paper introduces Avocado, a deep learning method that learns latent representations of human genomics data, enabling accurate imputation of unperformed experiments and improved performance on genomics tasks.

Ledidi designs genome edits to induce functional activity by optimizing predictions from a model, validated using Basenji for CTCF and JUND binding.

The paper screens random peptides to identify transcription activation domains and develops a deep learning model to predict their function and characteristics.

The paper introduces Avocado, a deep tensor factorization method that learns a latent representation of the human epigenome, improving imputation and performance on genomics tasks.

The paper uses Avocado to impute epigenomic and transcriptomic data across 400 human biosamples and 84 assays, improving protein binding predictions.

The paper profiles CRISPR/Cas9-induced mutational outcomes and develops a predictive model based on sequence context to guide DSB repair outcomes.

Rambutan, a convolutional network, calls high-confidence 1 kb Hi-C contacts from DNA sequence plus DNaseI signal across six cell types; the authors flag an error in contact calling.