Our new biophysically interpretable machine learning approach has been published by @NatureBiotech. This was a great collaborative effort. #ProBound

Thanks again to @tuuliel_lab for a great collaboration!.

Excited to share that our paper came out in Cell Genomics last week. We developed a model to estimate TF regulatory activity from GTEx RNA-seq data and mapped genetic variants (“aQTLs”) predictive of this virtual trait.

Our scripts and data are available at and We greatly appreciate the advice we got from members of the Bussemaker and Lappalainen labs. Feedback is welcome! (n/n).

Altogether, the set of computational tools we introduce here highlights the potential of genetic association studies for cellular endophenotypes based on a network-based multi-omic approach. (8/n).

The trait TF tends to have an association with the protein product of the closest gene. For instance, one of the plausible genetic determinants of RELA activity is an aQTL in the 3’-UTR of SMAD4, whose protein product has been reported to physically interact with RELA. (7/n)

We analyzed fine-mapped aQTLs in terms of functional categories, and found that at more stringent p-value cutoffs, the identified variants are significantly enriched for open chromatin as well as TF binding sites. (6/n)

We mapped aQTLs for 55 TFs across 49 tissues using genome-wide association analysis of inferred TF activity. Taking multiple testing Into account, at the most stringent cutoff of p<1E–11, we discover 111 aQTLs, at an FDR estimated to be around 25%. (5/n)

One initial biological finding that validated our approach is that the activity of HMGB2 as estimated from the GTEx data significantly declines with age, consistent with previous reports. Other tissues also show age-dependent HMGB2 activity. (4/n)

We fit a beta-binomial generalized linear model (GLM) at the level of *pairs* of neighboring genes to control for variation in local chromatin structure and other confounding effects. As the predictor in our model we use TF perturbation signatures from the ENCODE project. (3/n)

To map genetic variants that impact the expression of many genes simultaneously through a shared TF, we used an approach in which the protein-level regulatory activity of the TF is inferred from RNA-seq counts, and then genetically mapped as a molecular quantitative trait. (2/n)

Excited to share a new preprint by my PhD student @xiaoting_l from our wonderful collaboration with the @tuuliel_lab on genetic determinants of TF activity ("aQTLs"). (1/n).

The website that accompanies our paper provides sequence-to-affinity models for hundreds of transcription factors, with direct links to cross-platform validation results for each. #MotifCentral.

Our lab is looking for an outstanding experimental postdoc to help develop the next generation of biomolecular profiling in a joint project with @NShahLab. See for details. @Chemjobber #postdocpositions.

The Bussemaker Lab @Columbia is hiring talented programmers and postdocs to advance new ideas and ongoing projects in #RegulatoryGenomics, #ComputationalBiology, and #MachineLearning! Come join us!

Giving a #BiocNYC talk and hands-on R tutorial about #NoReadLeftBehind and related tools for building protein-DNA recognition models from high-throughput SELEX data this Friday at Rockefeller University, together with lab member Chaitanya Rastogi: