Identifying motif combinations with extreme expression
Year: 2026
Type: article
Abstract: ATLAS_data: Sequence-to-expression datasets and reproduction code for motif inference This repository accompanies the manuscript's sequence-to-expression analyses. It provides a fully traceable data pipeline — from each dataset's original public source, through the scripts that convert it to a uniform format, to the analysis-ready inputs and the code that reproduces the reported results. Contents (four pipeline stages): seq_exp_csv_sources/ — provenance of every raw download: upstream repository or GEO accession, exact files used, filtering applied, citations, and (where needed) small extraction scripts that convert the original release into a parseable CSV. from_csv_to_jld2/ — Julia scripts that parse those CSVs into uniform .jld2 files, each holding one SEQ2EXP_Dataset pairing sequences with their expression/activity labels. _data_jld2s/ — the resulting analysis-ready datasets, one folder per dataset, referenced directly by the analysis code. get_results/ — get_results.jl reads the datasets and reproduces the experimental results (per-dataset model metrics, inferred motifs, trained checkpoints, and motif visualizations). Datasets (eight, four readout types): promoter expression (yeast GPRA, E. coli cross-species, plant leaf/protoplast), alternative 5′ splice-site usage (Rosenberg A5SS, SPLIRENT), and 5′ UTR translation efficiency (human GSE114002, yeast GSE104252). Reproducibility: results are pinned to MotifInference v0.0.6, with a full Manifest.toml lock of every dependency. Built and run with Julia 1.12.2 and CUDA.jl v5.9.7. An NVIDIA CUDA-capable GPU is required to reproduce the results.
Authors: Zhang, Chaolin, Chu, Shane
Institution: Columbia University
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7166105545