Collaborative Completion of Transcription Factor Binding Profiles via Local Sensitive Unified Embedding

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• ChIP-seq data enables comparative studies of regulatory activities.
• Limited data availability restricts comprehensive integrative analyses.
• Existing imputation methods like ChromImpute struggle with scarce data.

Purpose of the Study:

• To address the limitations of ChromImpute in predicting ChIP-seq datasets with scarce data.
• To propose a novel imputation method, Local Sensitive Unified Embedding (LSUE).
• To enhance the accuracy and applicability of ChIP-seq data imputation for TF binding prediction.

Main Methods:

• LSUE fuses ChIP-seq data by mapping proteins, samples, and genomic positions into a unified Euclidean space.
• It formulates representation learning as a single optimization problem to capture overall data structure.
• Incorporates local sensitive low-rank regularization for improved performance.

Main Results:

• LSUE demonstrates improved performance over ChromImpute, especially with limited available data.
• The method accurately approximates experimentally measured signals, enhancing integrative analysis.
• Experimental evaluations on ENCODE TF ChIP-seq data validate LSUE's effectiveness.

Conclusions:

• LSUE offers a robust solution for ChIP-seq data imputation, overcoming limitations of previous methods.
• The unified embedding approach captures complex data relationships for accurate predictions.
• LSUE has the potential to significantly advance integrative genomics studies.