Transfer String Kernel for Cross-Context DNA-Protein Binding Prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate prediction of transcription factor binding sites (TFBS) is crucial for understanding gene regulation.
• Existing methods struggle with predicting TFBS in unannotated cellular contexts due to sample distribution shifts between source and target data.
• This limitation hinders the accurate identification of regulatory elements in novel biological systems.

Purpose of the Study:

• To develop a novel method, Transfer String Kernel (TSK), for accurate TFBS prediction in unannotated cellular contexts.
• To address the challenge of sample distribution shift in cross-context predictions.
• To improve the transferability of predictive models across different biological contexts.

Main Methods:

• Proposed Transfer String Kernel (TSK), a method employing knowledge transfer via cross-context sample adaptation.
• Utilized a discriminative mismatch string kernel to map sequence segments into a high-dimensional feature space.
• Implemented sample re-weighting in the high-dimensional space to align source and target context distributions.

Main Results:

• TSK demonstrated superior performance in TFBS identification across 14 different transcription factors (TFs) in a cross-organism setting.
• The method showed significant improvements, particularly for TFs with non-conserved binding sequences across contexts.
• TSK also proved generalizable, achieving state-of-the-art results on MHC peptide binding predictions.

Conclusions:

• TSK effectively overcomes the limitations of existing methods in cross-context TFBS prediction.
• The proposed approach offers a robust solution for identifying TFBS in unannotated cellular environments.
• TSK's generalizability highlights its potential for broader applications in bioinformatics and computational biology.