Athena: Automated Tuning of k-mer based Genomic Error Correction Algorithms using Language Models

MAC-ErrorReads: machine learning-assisted classifier for filtering erroneous NGS reads

BACKGROUND

The rapid advancement of next-generation sequencing (NGS) machines in terms of speed and affordability has led to the generation of a massive amount of biological data at the expense of data quality as errors become more prevalent. This introduces the need to utilize different approaches to detect and filtrate errors, and data quality assurance is moved from the hardware space to the software preprocessing stages.

RESULTS

We introduce MAC-ErrorReads, a novel Machine learning-Assisted Classifier designed for filtering Erroneous NGS Reads. MAC-ErrorReads transforms the erroneous NGS read filtration process into a robust binary classification task, employing five supervised machine learning algorithms. These models are trained on features extracted through the computation of Term Frequency-Inverse Document Frequency (TF_IDF) values from various datasets such as E. coli, GAGE S. aureus, H. Chr14, Arabidopsis thaliana Chr1 and Metriaclima zebra. Notably, Naive Bayes demonstrated robust performance across various datasets, displaying high accuracy, precision, recall, F1-score, MCC, and ROC values. The MAC-ErrorReads NB model accurately classified S. aureus reads, surpassing most error correction tools with a 38.69% alignment rate. For H. Chr14, tools like Lighter, Karect, CARE, Pollux, and MAC-ErrorReads showed rates above 99%. BFC and RECKONER exceeded 98%, while Fiona had 95.78%. For the Arabidopsis thaliana Chr1, Pollux, Karect, RECKONER, and MAC-ErrorReads demonstrated good alignment rates of 92.62%, 91.80%, 91.78%, and 90.87%, respectively. For the Metriaclima zebra, Pollux achieved a high alignment rate of 91.23%, despite having the lowest number of mapped reads. MAC-ErrorReads, Karect, and RECKONER demonstrated good alignment rates of 83.76%, 83.71%, and 83.67%, respectively, while also producing reasonable numbers of mapped reads to the reference genome.

CONCLUSIONS

This study demonstrates that machine learning approaches for filtering NGS reads effectively identify and retain the most accurate reads, significantly enhancing assembly quality and genomic coverage. The integration of genomics and artificial intelligence through machine learning algorithms holds promise for enhancing NGS data quality, advancing downstream data analysis accuracy, and opening new opportunities in genetics, genomics, and personalized medicine research.

Genome sequence assembly algorithms and misassembly identification methods

The sequence assembly algorithms have rapidly evolved with the vigorous growth of genome sequencing technology over the past two decades. Assembly mainly uses the iterative expansion of overlap relationships between sequences to construct the target genome. The assembly algorithms can be typically classified into several categories, such as the Greedy strategy, Overlap-Layout-Consensus (OLC) strategy, and de Bruijn graph (DBG) strategy. In particular, due to the rapid development of third-generation sequencing (TGS) technology, some prevalent assembly algorithms have been proposed to generate high-quality chromosome-level assemblies. However, due to the genome complexity, the length of short reads, and the high error rate of long reads, contigs produced by assembly may contain misassemblies adversely affecting downstream data analysis. Therefore, several read-based and reference-based methods for misassembly identification have been developed to improve assembly quality. This work primarily reviewed the development of DNA sequencing technologies and summarized sequencing data simulation methods, sequencing error correction methods, various mainstream sequence assembly algorithms, and misassembly identification methods. A large amount of computation makes the sequence assembly problem more challenging, and therefore, it is necessary to develop more efficient and accurate assembly algorithms and alternative algorithms.

CARE 2.0: reducing false-positive sequencing error corrections using machine learning

BACKGROUND

Next-generation sequencing pipelines often perform error correction as a preprocessing step to obtain cleaned input data. State-of-the-art error correction programs are able to reliably detect and correct the majority of sequencing errors. However, they also introduce new errors by making false-positive corrections. These correction mistakes can have negative impact on downstream analysis, such as k-mer statistics, de-novo assembly, and variant calling. This motivates the need for more precise error correction tools.

RESULTS

We present CARE 2.0, a context-aware read error correction tool based on multiple sequence alignment targeting Illumina datasets. In addition to a number of newly introduced optimizations its most significant change is the replacement of CARE 1.0's hand-crafted correction conditions with a novel classifier based on random decision forests trained on Illumina data. This results in up to two orders-of-magnitude fewer false-positive corrections compared to other state-of-the-art error correction software. At the same time, CARE 2.0 is able to achieve high numbers of true-positive corrections comparable to its competitors. On a simulated full human dataset with 914M reads CARE 2.0 generates only 1.2M false positives (FPs) (and 801.4M true positives (TPs)) at a highly competitive runtime while the best corrections achieved by other state-of-the-art tools contain at least 3.9M FPs and at most 814.5M TPs. Better de-novo assembly and improved k-mer analysis show the applicability of CARE 2.0 to real-world data.

CONCLUSION

False-positive corrections can negatively influence down-stream analysis. The precision of CARE 2.0 greatly reduces the number of those corrections compared to other state-of-the-art programs including BFC, Karect, Musket, Bcool, SGA, and Lighter. Thus, higher-quality datasets are produced which improve k-mer analysis and de-novo assembly in real-world datasets which demonstrates the applicability of machine learning techniques in the context of sequencing read error correction. CARE 2.0 is written in C++/CUDA for Linux systems and can be run on the CPU as well as on CUDA-enabled GPUs. It is available at https://github.com/fkallen/CARE .

Lerna: transformer architectures for configuring error correction tools for short- and long-read genome sequencing

BACKGROUND

Sequencing technologies are prone to errors, making error correction (EC) necessary for downstream applications. EC tools need to be manually configured for optimal performance. We find that the optimal parameters (e.g., k-mer size) are both tool- and dataset-dependent. Moreover, evaluating the performance (i.e., Alignment-rate or Gain) of a given tool usually relies on a reference genome, but quality reference genomes are not always available. We introduce Lerna for the automated configuration of k-mer-based EC tools. Lerna first creates a language model (LM) of the uncorrected genomic reads, and then, based on this LM, calculates a metric called the perplexity metric to evaluate the corrected reads for different parameter choices. Next, it finds the one that produces the highest alignment rate without using a reference genome. The fundamental intuition of our approach is that the perplexity metric is inversely correlated with the quality of the assembly after error correction. Therefore, Lerna leverages the perplexity metric for automated tuning of k-mer sizes without needing a reference genome.

RESULTS

First, we show that the best k-mer value can vary for different datasets, even for the same EC tool. This motivates our design that automates k-mer size selection without using a reference genome. Second, we show the gains of our LM using its component attention-based transformers. We show the model's estimation of the perplexity metric before and after error correction. The lower the perplexity after correction, the better the k-mer size. We also show that the alignment rate and assembly quality computed for the corrected reads are strongly negatively correlated with the perplexity, enabling the automated selection of k-mer values for better error correction, and hence, improved assembly quality. We validate our approach on both short and long reads. Additionally, we show that our attention-based models have significant runtime improvement for the entire pipeline-18[Formula: see text] faster than previous works, due to parallelizing the attention mechanism and the use of JIT compilation for GPU inferencing.

CONCLUSION

Lerna improves de novo genome assembly by optimizing EC tools. Our code is made available in a public repository at: https://github.com/icanforce/lerna-genomics .

CARE: context-aware sequencing read error correction

MOTIVATION

Error correction is a fundamental pre-processing step in many Next-Generation Sequencing (NGS) pipelines, in particular for de novo genome assembly. However, existing error correction methods either suffer from high false-positive rates since they break reads into independent k-mers or do not scale efficiently to large amounts of sequencing reads and complex genomes.

RESULTS

We present CARE-an alignment-based scalable error correction algorithm for Illumina data using the concept of minhashing. Minhashing allows for efficient similarity search within large sequencing read collections which enables fast computation of high-quality multiple alignments. Sequencing errors are corrected by detailed inspection of the corresponding alignments. Our performance evaluation shows that CARE generates significantly fewer false-positive corrections than state-of-the-art tools (Musket, SGA, BFC, Lighter, Bcool, Karect) while maintaining a competitive number of true positives. When used prior to assembly it can achieve superior de novo assembly results for a number of real datasets. CARE is also the first multiple sequence alignment-based error corrector that is able to process a human genome Illumina NGS dataset in only 4 h on a single workstation using GPU acceleration.

AVAILABILITYAND IMPLEMENTATION

CARE is open-source software written in C++ (CPU version) and in CUDA/C++ (GPU version). It is licensed under GPLv3 and can be downloaded at https://github.com/fkallen/CARE.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.