AI-Driven Genome Assembly & Variant Calling

AI and bioinformatics methods for more accurate, continuous genome assemblies and improved downstream analysis

Status: Published in G3: Genes, Genomes, Genetics (2018); methods extended with AI-based approaches First Authors: Edwin A. Solares & Mahul Chakraborty (co-first authors) Collaborators: Danny E. Miller, Shannon Kalsow, Kate Hall, Anoja G. Perera, J.J. Emerson, R. Scott Hawley

Overview

We develop AI and bioinformatics approaches to produce more accurate, continuous genome assemblies that enable better downstream analysis. Our foundational work demonstrated that reference-quality genomes could be assembled at a fraction of traditional costs, and ongoing efforts apply machine learning to improve assembly contiguity, error correction, and structural variant resolution.

The Problem

Fragmented or error-prone genome assemblies introduce systematic biases into every analysis built on top of them: gene annotation, variant calling, population genomics, and evolutionary comparisons all suffer when the underlying reference is incomplete.

  • Assembly gaps obscure repetitive and structurally complex regions where important functional elements reside
  • Misassemblies propagate errors into downstream analyses, leading to false variant calls and incorrect evolutionary inferences
  • Cost barriers historically limited high-quality assemblies to a handful of model organisms

Our Approach

Foundational Methods

We demonstrated that combining low-coverage long-read sequencing (PacBio, ~20-30X) with strategic use of related reference genomes and optimized assembly algorithms could produce reference-quality genomes at less than 10% of traditional costs.

AI and Computational Extensions

  • Machine learning for assembly quality assessment and error detection
  • Computational methods for resolving complex structural variants and repetitive regions
  • Automated pipelines integrating multiple data types (long reads, short reads, Hi-C) for maximally continuous assemblies

Key Results

Assembly Quality

  • Achieved reference-quality contiguity (N50 >1 Mb)
  • >98% gene completeness (BUSCO assessment)
  • Accurate representation of structural variants critical for downstream population and evolutionary analyses

Accessibility

  • 90% cost reduction compared to traditional methods
  • Total cost: <$2,000 per genome (vs. $20,000+ traditional)
  • Method adopted by dozens of labs worldwide across insects, plants, and fungi

Impact

  • Cited 75+ times in genome assembly literature
  • Contributed to the broader movement of making high-quality genomics accessible to resource-limited labs
  • Continuous, accurate assemblies from this work underpin downstream analyses across multiple ESB AI Lab research programs, including crop genomics, conservation genetics, and structural variant discovery

Publication

Solares, E.A., Chakraborty, M., Miller, D.E., et al. (2018). “Rapid Low-Cost Assembly of the Drosophila melanogaster Reference Genome Using Low-Coverage, Long-Read Sequencing.” G3: Genes, Genomes, Genetics, 8(10), 3143-3154.

References