Biology subjects

Yelick, K.

Publications and source records attributed to Yelick, K..

3 recordsLinked to original sources

bioRxiv2024

GenomeFace: a deep learning-based metagenome binner trained on 43,000 microbial genomes

Metagenomic binning, the process of grouping DNA sequences into taxonomic units, is critical for understanding the functions, interactions, and evolutionary dynamics of microbial communities. We propose a deep learning approach to binning using two neural networks, one based on composition and another on environmental abundance, dynamically weighting the contribution of each based on characteristics of the input data. Trained on over 43,000 prokaryotic genomes, our network for composition-based binning is inspired by metric learning techniques used for facial recognition. Using a task-specific, multi-GPU accelerated algorithm to cluster the embeddings produced by our network, our binner leverages marker genes observed to be universally present in nearly all taxa to grade and select optimal clusters of sequences from a hierarchy of candidates. We evaluate our approach on four simulated datasets with known ground truth. Our linear time integration of marker genes recovers more near complete genomes than state of the art but computationally infeasible solutions using them, while being over an order of magnitude faster. Finally, we demonstrate the scalability and acuity of our approach by testing it on three of the largest metagenome assemblies ever performed. Compared to other binners, we produced 47%-183% more near complete genomes. From these datasets, we find over the genomes of over 3000 new candidate species which have never been previously cataloged, representing a potential 4% expansion of the known bacterial tree of life.

bioinformatics↗

bioRxiv2020

GPU accelerated partial order multiple sequencealignment for long reads self-correction

As third generation sequencing technologies become more reliable and widely used to solve several genome-related problems, self-correction of long reads is becoming the preferred method to reduce the error rate of Pacific Biosciences and Oxford Nanopore long reads, that is now around 10-12%. Several of these self-correction methods rely on some form of Multiple Sequence Alignment (MSA) to obtain a consensus sequence for the original reads. In particular, error-correction tools such as RACON and CONSENT use Partial Order (PO) graph alignment to accomplish this task. PO graph alignment, which is computationally more expensive than optimal global pairwise alignment between two sequences, needs to be performed several times for each read during the error correction process. GPUs have proven very effective in accelerating several compute-intensive tasks in different scientific fields. We harnessed the power of these architectures to accelerate the error correction process of existing self-correction tools, to improve the efficiency of this step of genome analysis. In this paper, we introduce a GPU-accelerated version of the PO alignment presented in the POA v2 software library, implemented on an NVIDIA Tesla V100 GPU. We obtain up to 6.5x speedup compared to 64 CPU threads run on two 2.3 GHz 16-core Intel Xeon Processors E5-2698 v3. In our implementation we focused on the alignment of smaller sequences, as the CONSENT segmentation strategy based on k-mer chaining provides an optimal opportunity to exploit the parallel-processing power of GPUs. To demonstrate this, we have integrated our kernel in the CONSENT software. This accelerated version of CONSENT provides a speedup for the whole error correction step that ranges from 1.95x to 8.5x depending on the input reads.

bioinformatics↗

bioRxiv2018

BELLA: Berkeley Efficient Long-Read to Long-Read Aligner and Overlapper

Recent advances in long-read sequencing enable the characterization of genome structure and its intra- and inter-species variation at a resolution that was previously impossible. Detecting overlaps between reads is integral to many long-read genomics pipelines, such as de novo genome assembly. While longer reads simplify genome assembly and improve the contiguity of the reconstruction, current long-read technologies come with high error rates. We present Berkeley Long-Read to Long-Read Aligner and Overlapper (BELLA), a novel algorithm for computing overlaps and alignments via sparse matrix-matrix multiplication that balances the goals of recall and precision, performing well on both. We present a probabilistic model that demonstrates the feasibility of using short k-mers for detecting candidate overlaps. We then introduce a notion of reliable k-mers based on our probabilistic model. Combining reliable k-mers with our binning mechanism eliminates both the k-mer set explosion that would otherwise occur with highly erroneous reads and the spurious overlaps from k-mers originating in repetitive regions. Finally, we present a new method based on Chernoff bounds for separating true overlaps from false positives using a combination of alignment techniques and probabilistic modeling. Our methodologies aim at maximizing the balance between precision and recall. On both real and synthetic data, BELLA performs amongst the best in terms of F1 score, showing performance stability which is often missing for competitor software. BELLAs F1 score is consistently within 1.7% of the top entry. Notably, we show improved de novo assembly results on synthetic data when coupling BELLA with the Miniasm assembler.

bioinformatics↗