Boosting tissue-specific prediction of active cis-regulatory regions through deep learning and Bayesian optimization...

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Machine Learning in Biology

Background:

• Cis-regulatory regions (CRRs) are crucial non-coding DNA elements regulating gene transcription spatio-temporally.
• Genetic variants in CRRs are linked to pathogenicity, making their accurate identification vital for understanding human diseases.
• Machine learning (ML) methods are increasingly used to predict CRR activity, but challenges remain in tissue-specific identification.

Purpose of the Study:

• To compare the performance of two deep neural networks (DNNs) for predicting enhancer and promoter activity in specific cell lines.
• To investigate the impact of experimental setup, including Bayesian optimization for model selection and data rebalancing, on prediction accuracy.
• To evaluate the utility of sequence-only data versus epigenomic features for CRR activity prediction.

Main Methods:

• Comparison of a Feed-Forward Neural Network (FFNN) utilizing epigenomic features with a Convolutional Neural Network (CNN) using only genomic sequence data.
• Application of Bayesian optimization for automatic model selection to optimize learner performance.
• Exploration of various data rebalancing strategies to mitigate class imbalance issues in prediction tasks.

Main Results:

• Bayesian optimization significantly enhances the quality and performance of the predictive models.
• Data rebalancing strategies have a substantial impact on model prediction performance, with cautious application of test set rebalancing recommended to avoid over-optimistic results.
• CNNs, even when trained solely on genomic sequence data, achieve performance comparable to FFNNs that use epigenomic information.

Conclusions:

• Both sequence data and epigenomic information contain valuable content for predicting CRR activity.
• The study highlights the effectiveness of Bayesian optimization and data rebalancing techniques in improving ML model performance for CRR identification.
• Future research should consider integrating both sequence-based and epigenomic feature models to leverage complementary information for enhanced CRR activity prediction.