Identification of haploinsufficient genes from epigenomic data using deep forest

Area of Science:

• Genetics and Genomics
• Computational Biology
• Bioinformatics

Background:

• Haploinsufficiency, where one functional gene copy is insufficient, contributes to diseases like cancer and neurodevelopmental disorders.
• Existing computational methods for identifying haploinsufficient genes are hampered by study bias, experimental noise, and instability.
• There is a critical need for robust computational tools to accurately identify haploinsufficient genes.

Purpose of the Study:

• To develop an advanced computational model, HaForest, for the accurate identification of haploinsufficient genes.
• To address the limitations of current methods in detecting haploinsufficient genes, improving diagnostic and research capabilities.

Main Methods:

• Proposed a deep forest model (HaForest) integrating multiscale scanning for feature representation and a cascade forest structure.
• Employed Linear Discriminant Analysis for feature extraction and the LightGBM library to model complex dependencies among genes.
• Validated HaForest's performance against existing computational methods and deep learning algorithms using five epigenomic datasets.

Main Results:

• HaForest demonstrated superior performance in identifying haploinsufficient genes compared to established computational and deep learning approaches.
• The model effectively extracts contextual representations and exploits gene dependencies, leading to enhanced identification accuracy.
• Results highlight HaForest's unique and complementary capabilities in the field of haploinsufficiency gene identification.

Conclusions:

• HaForest offers a significant advancement in the computational identification of haploinsufficient genes.
• The model's robust performance provides a valuable tool for research into haploinsufficiency-related diseases.
• A standalone tool is available for broader scientific application and validation.