Incorporating deep learning TF models into SNEEP

Project Description

In our lab, we developed a statistical approach (SNEEP, see project site) that identifies regulatory single nucleotide variants (SNVs) affecting transcription factor (TF) binding. Currently, the approach uses position weight matrices (PWMs) as TF models. However, these models have disadvantages, and the community has been investigating more reliable models based on deep learning, such as BPNet. Recently, the TF database JASPAR extended its set of available motifs by adding a large collection of TF-specific BPNet models, which can be used as an alternative to the classical PWM approach.

This project focuses on improving and adapting our SNEEP approach by incorporating these deep learning models as TF models.

Research Goals

• Explore how to derive a differential TF binding score and p-value using BPNet models per TF. Investigate whether the SNEEP statistic is still applicable. If not, figure out which changes need to be made.
• Analyze how deletions and insertions can be handled
• Adapt the SNEEP software accordingly
• Test the new implementation
• Evaluate and compare the results to the previous version of SNEEP

Prerequisites

• C++ programming or motivation to learn it
• Basic knowlegde of genomics is helpful, but not required

How to Apply

Send an Email with a short motivation to the corresponding supervisor n.baumgarten|em.uni-frankfurt.de.