Helixer

Helixer applies deep learning to generate base-pair-wise probability annotations of genes across diverse eukaryotic genomes to improve de novo gene annotation and cross-species generalization.

Key Features:

• Cross-Species Generalization: Eliminates dependency on closely related gene models and performs annotation from DNA sequences alone across species.
• Deep Learning Architecture: Employs a fundamentally new deep learning architecture to enhance learning capacity for gene prediction tasks.
• Base-Wise Annotation: Produces base pair-wise probabilities rather than comprehensive gene models, providing fine-grained per-base likelihoods.
• Scalability Across Genomes: Evaluated using a single vertebrate model for annotation of 186 animal genomes and a separate land plant model for 51 plant genomes, demonstrating scalability and reduced sensitivity to genome length relative to state-of-the-art tools.
• Post-Processing Techniques: Incorporates two novel post-processing techniques to refine and strengthen gene annotations.
• RNA-Seq Based Validation: Predictions have been validated through RNA-Seq comparisons to assess concordance with transcriptomic data.

Scientific Applications:

• Comparative Genomics: Enables cross-species gene annotation for comparative genomic analyses.
• Evolutionary Biology: Supports analysis of gene evolution across diverse eukaryotic lineages.
• Functional Genomics: Facilitates identification and characterization of genes and gene structures for functional studies.
• Novel Gene Discovery: Aids discovery of previously unannotated genes and exploration of genomic diversity.
• Gene Function Inference: Improves annotation quality used to infer gene function from genomic sequence and expression data.

Methodology:

Train a unified deep learning model on diverse genomic data from multiple species; evaluate a vertebrate-trained model on 186 animal genomes and a land plant-trained model on 51 plant genomes; apply two post-processing techniques to refine predictions; and validate predictions via RNA-Seq comparisons.