CoNSEPT

CoNSEPT predicts gene expression from enhancer DNA sequences using convolutional neural networks to model cis- and trans-regulatory influences and was developed to analyze neuroectodermal enhancers in Drosophila.

Key Features:

• Convolutional Neural Network Architecture: CoNSEPT employs convolutional neural networks that learn enhancer "grammar" from sequence to capture complex sequence-function relationships indicative of regulatory potential.
• Comprehensive Input Parameters: Inputs include enhancer sequences, transcription factor (TF) levels across trans conditions, TF motifs represented as position weight matrices (PWMs), and known TF–TF interactions.
• Mechanistic Insights: The model yields interpretable predictions relating enhancer sequence features to mechanisms such as cooperative activation, short-range repression, and distance-independent repression.
• Model Comparison and Validation: CoNSEPT was developed alongside biophysical and other machine-learning models to enable rigorous comparison and mechanistic inference.

Scientific Applications:

• Interpreting Non-coding Variants: By predicting how enhancer sequences affect expression, CoNSEPT can be applied to interpret functional consequences of non-coding genetic variants.
• Transcriptomic Variation Studies: It supports studies investigating sources and mechanisms of transcriptomic variation across biological contexts.
• Mechanistic Biology Research: It facilitates testing hypotheses about enhancer function, including cooperative activation and repression mechanisms.

Methodology:

CoNSEPT uses convolutional neural networks trained to map enhancer sequences to expression, taking as inputs TF levels, PWMs, and known TF–TF interactions, and was evaluated in comparison with biophysical and other machine-learning models.