SemanticCAP: Chromatin Accessibility Prediction Enhanced by Features Learning from a Language Model.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Chromatin accessibility influences drug-DNA interactions and gene expression, impacting drug resistance.
• Experimental methods for measuring chromatin accessibility are costly and time-intensive.
• Current computational models often overlook the contextual information within gene sequences.

Purpose of the Study:

• To develop an advanced computational model, SemanticCAP, for predicting chromatin accessibility.
• To leverage gene language models for effective representation of genomic sequence context.
• To improve the accuracy and efficiency of chromatin accessibility prediction.

Main Methods:

• Developed SemanticCAP, integrating a gene language model to capture sequence context.
• Implemented SFA and SFC methods for seamless feature fusion within the chromatin accessibility model.
• Evaluated SemanticCAP against established methods like DeepSEA, gkm-SVM, and k-mer using public benchmarks.

Main Results:

• SemanticCAP demonstrated superior performance compared to existing computational models.
• Achieved a maximum improvement of 1.25% in Area Under the Receiver Operating Characteristic Curve (auROC).
• Showcased a maximum improvement of 2.41% in Area Under the Precision-Recall Curve (auPRC).

Conclusions:

• SemanticCAP offers a more effective approach to predicting chromatin accessibility by utilizing gene sequence context.
• The model's enhanced performance has implications for understanding drug-DNA interactions and gene regulation.
• SemanticCAP provides a valuable computational tool for genomic research, potentially reducing experimental costs and time.