Motif-based model of transcription predicts effects of sequence variants in AR enhancers and reveals distinct functions for AR-associated transcription factors

A sophisticated computational model has been developed to predict the impact of genetic variations on the activity of androgen receptor (AR) enhancers, particularly in the context of prostate cancer. This model, which is rooted in biophysical principles and offers interpretability, was trained on data from STARR-seq experiments conducted in LNCaP cells.

By dividing the analysis into two separate components - one focused on the effect of androgen induction on enhancer activity and the other on the baseline enhancer activity - the researchers were able to categorize transcription factors (TFs) into three distinct functional groups: hormone-dependent drivers, constitutive activators, and dual-role factors that affect both types of activity.

These classifications shed light on the complex interplay between AR and co-activators, revealing that inducibility is linked not only to the presence of AR and co-activator motifs but also to a reduced presence of constitutive activators that might otherwise limit the overall enhancer output. The model's predictive power was confirmed by comparing its predictions to an independent dataset derived from saturation mutagenesis studies on 40 AR enhancers.

This comparison revealed that the model could accurately forecast the effects of single-base mutations with an area under the curve (AUC) of 0.76. Furthermore, independent application of the model to genetic risk variants identified through genome-wide association studies (GWAS) in AR binding site regions yielded promising results.

The model successfully prioritized four specific candidate variants that were predicted to decrease the ratio of dihydrotestosterone (DHT) or ethyl alcohol (EtOH) enhancer activity at these loci, offering a potential avenue for further investigation into the genetic factors contributing to prostate cancer risk.