DeepConsensus

open_source

DeepConsensus uses gap-aware sequence transformers to correct errors in PacBio Circular Consensus Sequencing data, increasing the yield of high-quality reads.

Data & Analytics

AI Models & Infrastructure

Research & Education

About

DeepConsensus is a Google-developed open-source bioinformatics tool that applies deep learning — specifically gap-aware sequence transformers — to improve the accuracy of PacBio Circular Consensus Sequencing (CCS) reads. Long-read sequencing technologies like PacBio CCS generate valuable genomic data but are prone to systematic errors; DeepConsensus addresses this by training a transformer model to identify and correct these errors, resulting in a greater yield of high-quality reads across various chemistries and read length distributions. The tool is available as a Python package installable via pip and also provides a Dockerfile for containerized deployments. It supports GPU acceleration for faster inference and includes guidance for sharding and parallelizing workloads across large SMRT Cell datasets. DeepConsensus has been benchmarked on three full SMRT Cells with different chemistries, demonstrating measurable improvements in read quality and yield. Primary users include genomics researchers, bioinformaticians, and computational biology teams working in academic, clinical, or industrial settings. DeepConsensus integrates into existing sequencing pipelines as a post-processing step after CCS generation, requiring users to follow specific CCS settings for optimal results. The project is licensed under BSD-3-Clause and actively maintained on GitHub by Google's research team.

Key Features

Gap-Aware Sequence Transformers: Uses a specialized transformer architecture designed to handle the alignment gaps inherent in CCS data, enabling precise error identification and correction.
High-Quality Read Yield Improvement: Demonstrated improvements in the percentage of high-quality reads across multiple PacBio SMRT Cell chemistries and read length distributions.
Sharding and Parallelization Support: Built-in guidance and tooling for sharding large datasets and running parallel inference jobs to efficiently process full SMRT Cell outputs.
GPU Acceleration: Supports GPU-accelerated inference via dedicated install scripts, dramatically reducing processing time for large genomic datasets.
Docker and pip Deployment: Available as a pip-installable Python package and as a Dockerfile, making it easy to integrate into existing bioinformatics pipelines and cloud environments.

Use Cases

Improving the accuracy and yield of PacBio CCS reads in whole-genome sequencing projects before downstream variant calling.
Post-processing CCS data in clinical genomics pipelines to maximize high-quality read output from expensive SMRT Cell runs.
Benchmarking and evaluating PacBio chemistry versions by quantifying quality improvements across different read length distributions.
Integrating into HPC or cloud-based bioinformatics workflows via Docker for scalable, reproducible sequencing data processing.
Supporting genomics research teams that need to extract maximum value from long-read sequencing data with limited sample material.

Pros

Open Source with Permissive License: Released under BSD-3-Clause by Google, allowing free use, modification, and integration into both academic and commercial workflows.
Backed by Google Research: Developed and maintained by Google, ensuring rigorous benchmarking, ongoing updates, and scientific credibility.
Measurable Quality Gains: Published yield metrics across three full SMRT Cells demonstrate concrete, reproducible improvements in read quality over standard CCS output.
Flexible Deployment Options: Supports both pip installation and Docker containerization, with GPU support, making it adaptable to diverse compute environments including HPC clusters and cloud platforms.

Cons

PacBio CCS Only: Designed exclusively for PacBio Circular Consensus Sequencing data; not compatible with other long-read (e.g., Oxford Nanopore) or short-read sequencing technologies.
Requires Bioinformatics Expertise: Setup and optimal use require familiarity with sequencing pipelines, CCS configuration parameters, and command-line tools — not suitable for non-technical users.
Computationally Intensive: Processing full SMRT Cell datasets demands significant compute resources; GPU hardware is recommended for practical turnaround times.

Frequently Asked Questions

DeepConsensus is designed specifically for Pacific Biosciences (PacBio) Circular Consensus Sequencing (CCS) data. It acts as a post-processing step after CCS generation to further correct residual errors in the reads.

Yes. DeepConsensus is fully open source and released under the BSD-3-Clause license, allowing free use for both academic and commercial purposes.

You can install DeepConsensus via pip or use the provided Dockerfile. The repository includes a quick start guide covering installation, basic usage, and how to shard and parallelize runs for large datasets.

GPU usage is not strictly required, but it is strongly recommended. The repository includes a dedicated GPU installation script (install-gpu.sh) to enable faster model inference.

DeepConsensus applies a gap-aware sequence transformer model trained on aligned subreads to detect and correct errors that remain after standard CCS processing, increasing the proportion of reads meeting high-quality thresholds (e.g., Q20, Q30).