Biology subjects

Coleman, B.

Publications and source records attributed to Coleman, B..

5 recordsLinked to original sources

bioRxiv2022

Node-degree aware edge sampling mitigates inflated classification performance in biomedical graph representation learning

Graph representation learning is a family of related approaches that learn low-dimensional vector representations of nodes and other graph elements called embeddings. Embeddings approximate characteristics of the graph and can be used for a variety of machine-learning tasks such as novel edge prediction. For many biomedical applications, partial knowledge exists about positive edges that represent relationships between pairs of entities, but little to no knowledge is available about negative edges that represent the explicit lack of a relationship between two nodes. For this reason, classification procedures are forced to assume that the vast majority of unlabeled edges are negative. Existing approaches to sampling negative edges for training and evaluating classifiers do so by uniformly sampling pairs of nodes. We show here that this sampling strategy typically leads to sets of positive and negative edges with imbalanced edge degree distributions. Using representative homogeneous and heterogeneous biomedical knowledge graphs, we show that this strategy artificially inflates measured classification performance. We present a degree-aware node sampling approach for sampling negative edge examples that mitigates this effect and is simple to implement.

bioinformatics↗

bioRxiv2021

Supervised learning with word embeddings derived from PubMed captures latent knowledge about protein kinases and cancer

Inhibiting protein kinases (PKs) that cause cancers has been an important topic in cancer therapy for years. So far, almost 8% of more than 530 PKs have been targeted by FDA-approved medications and around 150 protein kinase inhibitors (PKIs) have been tested in clinical trials. We present an approach based on natural language processing and machine learning to the relations between PKs and cancers, predicting PKs whose inhibition would be efficacious to treat a certain cancer. Our approach represents PKs and cancers as semantically meaningful 100-dimensional vectors based on co-occurrence patterns in PubMed abstracts. We use information about phase I-IV trials in ClinicalTrials.gov to construct a training set for random forest classification. In historical data, associations between PKs and specific cancers could be predicted years in advance with good accuracy. Our model may be a tool to predict the relevance of inhibiting PKs with specific cancers.

bioinformatics↗

bioRxiv2019

Diversified RACE Sampling on Data Streams Applied to Metagenomic Sequence Analysis

The rise of whole-genome shotgun sequencing (WGS) has enabled numerous breakthroughs in large-scale comparative genomics research. However, the size of genomic datasets has grown exponentially over the last few years, leading to new challenges for traditional streaming algorithms. Modern petabyte-sized genomic datasets are difficult to process because they are delivered by high-throughput data streams and are difficult to store. As a result, many traditional streaming problems are becoming increasingly relevant. One such problem is the task of constructing a maximally diverse sample over a data stream. In this regime, complex sampling procedures are not possible due to the overwhelming data generation rate. In theory, the best diversity sampling methods are based on a simple greedy algorithm that compares the current sequence with a large pool of sampled sequences and decides whether to accept or reject the sequence. While these methods are elegant and optimal, they are largely confined to the theoretical realm because the greedy procedure is too slow in practice. While there are many methods to identify common elements in data streams efficiently, fast and memory-efficient diversity sampling remains a challenging and fundamental data streaming problem with few satisfactory solutions. In this work, we bridge the gap with RACE sampling, an online algorithm for diversified sampling. Unlike random sampling, which samples uniformly, RACE selectively accepts samples from streams that lead to higher sequence diversity. At the same time, RACE is as computationally efficient as random sampling and avoids pairwise similarity comparisons between sequences. At the heart of RACE lies an efficient lookup array constructed using locality-sensitive hashing (LSH). Our theory indicates that an accept/reject procedure based on LSH lookups is sufficient to obtain a highly diverse subsample. We provide rigorous theoretical guarantees for well-known biodiversity indices and show that RACE can nearly double the Shannon and Simpson indices of a genetic sample in practice, all while using the same resources as random sampling. We also compare RACE against Diginorm and coreset-based diversity sampling methods and find that RACE is faster and more memory efficient. Our algorithm is straightforward to implement, easy to parallelize, and fast enough to keep pace with the overwhelming data generation rates. We expect that as DNA sequence data streams become more mainstream and faster, RACE will become an essential component for many applications.1

bioinformatics↗

bioRxiv2017

The ERK MAPK pathway modulates Gq-dependent locomotion in Caenorhabditis elegans

The heterotrimeric G protein Gq regulates neuronal activity through distinct downstream effector pathways. In addition to the canonical Gq effector phospholipase C{beta}, the small GTPase Rho was recently identified as a conserved effector of Gq. To identify additional molecules important for Gq signaling in neurons, we performed a forward genetic screen in the nematode Caenorhabditis elegans for suppressors of the hyperactivity and exaggerated waveform of an activated Gq mutant. We isolated two mutations affecting the MAP kinase scaffold protein KSR-1 and found that KSR-1 modulates locomotion downstream of or in parallel to the Gq-Rho pathway. Through epistasis experiments, we found that the core ERK MAPK cascade is required for Gq-Rho regulation of locomotion, but that the canonical ERK activator LET-60/Ras may not be required. Through neuron-specific rescue experiments, we found that the ERK pathway functions in head acetylcholine neurons to control Gq-dependent locomotion. Additionally, expression of activated LIN-45/Raf in head acetylcholine neurons is sufficient to cause an exaggerated waveform phenotype and hypersensitivity to the acetylcholinesterase inhibitor aldicarb, similar to an activated Gq mutant. Taken together, our results suggest that the ERK MAPK pathway modulates the output of Gq-Rho signaling to control locomotion behavior in C. elegans.

genetics↗

bioRxiv2017

SARNAclust: Semi-Automatic Detection Of RNA Protein Binding Motifs From Immunoprecipitation Data

RNA-protein binding is critical to gene regulation, controlling fundamental processes including splicing, translation, localization and stability, and aberrant RNA-protein interactions are known to play a role in a wide variety of diseases. However, molecular understanding of RNA-protein interactions remains limited, and in particular identification of the RNA motifs that bind proteins has long been a difficult problem. To address this challenge, we have developed a novel semi-automatic algorithm, SARNAclust, to computationally identify combined structure/sequence motifs from immunoprecipitation data. SARNAclust is, to our knowledge, the first unsupervised method that can identify RNA motifs at full structural resolution while also being able to simultaneously deconvolve multiple motifs. SARNAclust makes use of a graph kernel to evaluate similarity between sequence/structure objects, and provides the ability to isolate the impact of specific features through the bulge graph formalism. SARNAclust includes a key method for predicting RNA secondary structure at CLIP peaks, RNApeakFold, which we have verified to be effective on synthetic motif data. We applied SARNAclust to 30 ENCODE eCLIP datasets, identifying known motifs and novel predictions. Notably, we predicted a new motif for the protein ILF3 similar to that for the splicing factor hnRNPC, providing evidence for interaction between these two proteins. To validate our predictions, we performed a directed RNA bind-n-seq assay for two proteins: ILF3 and SLBP, in each case revealing the effectiveness of SARNAclust in predicting RNA sequence and structure elements important to protein binding. Availability: https://github.com/idotu/SARNAclust

bioinformatics↗