Probabilistic Inference on Multiple Normalized Signal Profiles from Next Generation Sequencing: Transcription Factor...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Massive amounts of chromatin immunoprecipitation sequencing (ChIP-Seq) data are available, measuring genome-wide DNA-binding protein occupancy.
• Understanding gene regulation requires analyzing multiple ChIP-Seq profiles to decipher transcription causalities.

Purpose of the Study:

• To develop probabilistic models for inferring genome-wide ChIP-Seq signal profiles.
• To improve the understanding of gene transcription regulation by DNA-binding proteins.

Main Methods:

• Proposed SignalRanker, a probabilistic model assuming conditional independence of DNA-binding protein occupancies.
• Developed FullSignalRanker, a more advanced probabilistic model that does not assume conditional independence.
• Compared SignalRanker and FullSignalRanker against existing methods using ENCODE ChIP-Seq datasets.

Main Results:

• FullSignalRanker demonstrated superior performance in recovering signal ranks on promoter and enhancer regions.
• FullSignalRanker achieved the best performance in peak sequence classification.
• Both methods showed strong regression and classification abilities.

Conclusions:

• FullSignalRanker is a highly effective method for analyzing ChIP-Seq data, particularly for promoter and enhancer regions.
• The developed models are valuable tools for the analysis of next-generation sequencing data in genomics.
• FullSignalRanker program is publicly available for researchers.