deepSimDEF

deepSimDEF estimates functional similarity between genes by using a deep neural network to learn low-dimensional vector embeddings from Gene Ontology (GO) annotations of gene products such as RNAs and proteins for comparative genomic analyses.

Key Features:

• Deep Learning Framework: Employs a deep neural network that transforms GO annotations into low-dimensional embedding vector representations for GO terms and gene products.
• Automatic Feature Learning: Automatically learns features to quantify functional similarity without relying on hand-engineered metrics.
• Versatile Application: Handles single- and multi-channel data inputs for diverse genomic datasets.
• Performance Superiority: Demonstrated more than 5–10% improved performance on yeast and human reference datasets across protein-protein interactions, gene co-expression, and sequence homology tasks.

Scientific Applications:

• Protein-Protein Interaction Analysis: Predicts functional similarity from GO annotations to infer potential interactions between proteins.
• Gene Co-expression Studies: Assesses functional similarity to support identification of co-regulated genes from expression data.
• Sequence Homology Investigations: Provides nuanced functional relationship estimates to augment sequence homology and evolutionary analyses.

Methodology:

Transforms GO annotations into vector representations by learning low-dimensional embeddings for GO terms and gene products and computes functional similarity between gene pairs using these learned vectors.