Biology subjects

Schreiber, J.

Publications and source records attributed to Schreiber, J..

31 records · Page 2Linked to original sources

bioRxiv2020

Learning a latent representation of human genomics using Avocado

In the past decade, the use of high-throughput sequencing assays has allowed researchers to experimentally acquire thousands of functional measurements for each basepair in the human genome. Despite their value, these measurements are only a small fraction of the potential experiments that could be performed while also being too numerous to easily visualize or compute on. In a recent pair of publications, we address both of these challenges with a deep neural network tensor factorization method, Avocado, that compresses these measurements into dense, information-rich representations. We demonstrate that these learned representations can be used to impute with high accuracy the output of experimental assays that have not yet been performed and that machine learning models that leverage these representations outperform those trained directly on the functional measurements on a variety of genomics tasks. The code is publicly available at https://github.com/jmschrei/avocado.

bioinformatics↗

bioRxiv2020

Ledidi: Designing genome edits that induce functional activity

The development of modern genome editing tools has enabled researchers to make edits with high precision, but has left unsolved the problem of designing these edits. We propose Ledidi, an approach that treats the design of genomic edits as an optimization problem where the goal is to produce the desired output from a predictive model. The discrete nature of biological sequences makes direct optimization challenging, but we overcome this by using the Gumbel-Softmax reparameterization trick. We validate Ledidi by pairing it with the Basenji model, which makes predictions for thousands of functional profiles, and designing edits that affect CTCF binding and induce cell type-specific binding of JUND.

bioinformatics↗

bioRxiv2020

Submodular sketches of single-cell RNA-seq measurements

Analyzing and sharing massive single-cell RNA-seq data sets can be facilitated by creating a "sketch" of the data--a selected subset of cells that accurately represent the full data set. Using an existing benchmark, we demonstrate the utility of submodular optimization in efficiently creating high quality sketches of scRNA-seq data.

bioinformatics↗

bioRxiv2020

Age-related changes of Peak width Skeletonized Mean Diffusivity (PSMD) across the adult life span: a multi-cohort study

Parameters of water diffusion in white matter derived from diffusion-weighted imaging (DWI), such as fractional anisotropy (FA), mean, axial, and radial diffusivity (MD, AD and RD), and more recently, peak width of skeletonized mean diffusivity (PSMD), have been proposed as potential markers of normal and pathological brain ageing. However, their relative evolution over the entire adult lifespan in healthy individuals remains partly unknown during early and late adulthood, and particularly for the PSMD index. Here, we gathered and meta-analyzed cross-sectional diffusion tensor imaging (DTI) data from 10 population-based cohort studies in order to establish the time course of white matter water diffusion phenotypes from post-adolescence to late adulthood. DTI data were obtained from a total of 20,005 individuals aged 18.1 to 92.6 years and analyzed with the same pipeline for computing DTI metrics. For each individual MD, AD, RD, and FA mean values were computed over their FA volume skeleton, PSMD being calculated as the 90% peak width of the MD values distribution across the FA skeleton. Mean values of each DTI metric were found to strongly vary across cohorts, most likely due to major differences in DWI acquisition protocols as well as pre-processing and DTI model fitting. However, age effects on each DTI metric were found to be highly consistent across cohorts. RD, MD and AD variations with age exhibited the same U-shape pattern, first slowly decreasing during post-adolescence until the age of 30, 40 and 50, respectively, then progressively increasing until late life. FA showed a reverse profile, initially increasing then continuously decreasing, slowly until the 70s, then sharply declining thereafter. By contrast, PSMD constantly increased, first slowly until the 60s, then more sharply. These results demonstrate that, in the general population, age affects PSMD in a manner different from that of other DTI metrics. The constant increase in PSMD throughout the entire adult life, including during post-adolescence, indicates that PSMD could be an early marker of the ageing process.

neuroscience↗

bioRxiv2019

A high-throughput screen for transcription activation domains reveals their sequence characteristics and permits reliable prediction by deep learning

Transcription activation domains (ADs) are encoded by a wide range of seemingly unrelated amino acid sequences, making it difficult to recognize features that permit their dynamic behavior, fuzzy interactions and target specificity. We screened a large set of random 30-mer peptides for AD function and trained a deep neural network (ADpred) on the AD-positive and negative sequences. ADpred correctly identifies known ADs within protein sequences and accurately predicts the consequences of mutations. We show that functional ADs are (1) located within intrinsically disordered regions with biased amino acid composition, (2) contain clusters of hydrophobic residues near acidic side chains, (3) are enriched or depleted for particular dipeptide sequences, and (4) have higher helical propensity than surrounding regions. Taken together, our findings fit the model of "fuzzy" binding through hydrophobic protein-protein interfaces, where activator-coactivator binding takes place in a dynamic hydrophobic environment rather than through combinations of sequence-specific interactions.

molecular biology↗

bioRxiv2019

Zero-shot imputations across species are enabled through joint modeling of human and mouse epigenomics

Recent large-scale efforts to characterize functional activity in human have produced thousands of genome-wide experiments that quantify various forms of biochemistry, such as histone modifications, protein binding, transcription, and chromatin accessibility. Although these experiments represent a small fraction of the possible experiments that could be performed, they also make human more comprehensively characterized than any other species. We propose an extension to the imputation approach Avocado that enables the model to leverage genome alignments and the large number of human genomics data sets when making imputations in other species. We found that not only does this extension result in improved imputation of mouse functional experiments, but that the extended model is able to make accurate imputations for protein binding assays that have been performed in human but not in mouse. This ability to make "zero-shot" imputations greatly increases the utility of such imputation approaches and enables comprehensive imputations to be made for species even when experimental data are sparse. CCS CONCEPTS* Computing methodologies [->] Neural networks; Factorization methods; * Applied computing [->] Bioinformatics; Genomics. ACM Reference FormatJacob Schreiber, Deepthi Hegde, and William Noble. 2020. Zero-shot imputations across species are enabled through joint modeling of human and mouse epigenomics. In ACM-BCB 2020: 11th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics, Sept 21-24, 2020, Virtual. ACM, New York, NY, USA, 9 pages. https://doi.org/10.1145/1122445.1122456

bioinformatics↗

bioRxiv2019

Prioritizing transcriptomic and epigenomic experiments by using an optimization strategy that leverages imputed data

Successful science often involves not only performing experiments well, but also choosing well among many possible experiments. In a hypothesis generation setting, choosing an experiment well means choosing an experiment whose results are interesting or novel. In this work, we formalize this selection procedure in the context of genomics and epigenomics data generation. Specifically, we consider the task faced by a scientific consortium such as the National Institutes of Health ENCODE Consortium, whose goal is to characterize all of the functional elements in the human genome. Given a list of possible cell types or tissue types (\"biosamples\") and a list of possible high throughput sequencing assays, we ask \"Which experiments should ENCODE perform next?\" We demonstrate how to represent this task as an optimization problem, where the goal is to maximize the information gained in each successive experiment. Compared with previous work that has addressed a similar problem, our approach has the advantage that it can use imputed data to tailor the selected list of experiments based on data collected previously by the consortium. We demonstrate the utility of our proposed method in simulations, and we provide a general software framework, named Kiwano, for selecting genomic and epigenomic experiments.

genomics↗

bioRxiv2019

Completing the ENCODE3 compendium yields accurate imputations across a variety of assays and human biosamples

MotivationRecent efforts to describe the human epigenome have yielded thousands of uniformly processed epigenomic and transcriptomic data sets. These data sets characterize a rich variety of biological activity in hundreds of human cell lines and tissues ("biosamples"). Understanding these data sets, and specifically how they differ across biosamples, can help explain many cellular mechanisms, particularly those driving development and disease. However, due primarily to cost, the total number of assays that can be performed is limited. Previously described imputation approaches, such as Avocado, have sought to overcome this limitation by predicting genome-wide epigenomics experiments using learned associations among available epigenomic data sets. However, these previous imputations have focused primarily on measurements of histone modification and chromatin accessibility, despite other biological activity being crucially important. ResultsWe applied Avocado to a data set of 3,814 tracks of data derived from the ENCODE compendium, spanning 400 human biosamples and 84 assays. The resulting imputations cover measurements of chromatin accessibility, histone modification, transcription, and protein binding. We demonstrate the quality of these imputations by comprehensively evaluating the models predictions and by showing significant improvements in protein binding performance compared to the top models in an ENCODE-DREAM challenge. Additionally, we show that the Avocado model allows for efficient addition of new assays and biosamples to a pre-trained model, achieving high accuracy at predicting protein binding, even with only a single track of training data. AvailabilityTutorials and source code are available under an Apache 2.0 license at https://github.com/jmschrei/avocado. Contactwilliam-noble@uw.edu or jmschr@cs.washington.edu

bioinformatics↗

bioRxiv2019

A pitfall for machine learning methods aiming to predict across cell types

Machine learning models to predict phenomena such as gene expression, enhancer activity, transcription factor binding, or chromatin conformation are most useful when they can generalize to make accurate predictions across cell types. In this situation, a natural strategy is to train the model on experimental data from some cell types and evaluate performance on one or more held-out cell types. In this work, we show that, when the training set contains examples derived from the same genomic loci across multiple cell types, then the resulting model can be susceptible to a particular form of bias related to memorizing the average activity associated with each genomic locus. Consequently, the trained model may appear to perform well when evaluated on the genomic loci that it was trained on but tends to perform poorly on loci that it was not trained on. We demonstrate this phenomenon by using epigenomic measurements and nucleotide sequence to predict gene expression and chromatin domain boundaries, and we suggest methods to diagnose and avoid the pitfall. We anticipate that, as more data and computing resources become available, future projects will increasingly risk suffering from this issue.

bioinformatics↗

bioRxiv2018

A phylogeography of the second plague pandemic revealed through the analysis of historical Y. pestis genomes

The second plague pandemic (14th - 18th century AD), caused by the bacterium Yersinia pestis, is infamous for its initial wave, the Black Death (1346-1353 AD), and its repeated scourges in Europe and the vicinity until the Early Modern Era. Here, we report 32 ancient Y. pestis genomes spanning the 14th to 17th century AD through the analysis of human remains from nine European archaeological sites. Our data support an initial entry of the bacterium from Eastern Europe and the absence of genetic diversity during the Black Death as well as low diversity during local outbreaks thereafter. Moreover, analysis of post-Black Death genomes shows the diversification of a Y. pestis lineage into multiple genetically distinct clades that may have given rise to more than one disease reservoir in, or close to, Europe. Finally, we show the loss of a genomic region that includes virulence-associated genes in strains associated with late stages of the second plague pandemic (17th - 18th century AD). This deletion could not be detected in extant strains within our modern dataset, though it was identified in a today-extinct lineage associated with the first plague pandemic (6th - 8th century AD), suggesting convergent evolution during both pandemic events.

evolutionary biology↗

bioRxiv2018

Massively parallel profiling and predictive modeling of the outcomes of CRISPR/Cas9-mediated double-strand break repair

Non-homologous end-joining (NHEJ) plays an important role in double-strand break (DSB) repair of DNA. Recent studies have shown that the error patterns of NHEJ are strongly biased by sequence context, but these studies were based on relatively few templates. To investigate this more thoroughly, we systematically profiled [~]1.16 million independent mutational events resulting from CRISPR/Cas9-mediated cleavage and NHEJ-mediated DSB repair of 6,872 synthetic target sequences, introduced into a human cell line via lentiviral infection. We find that: 1) insertions are dominated by 1 bp events templated by sequence immediately upstream of the cleavage site, 2) deletions are predominantly associated with microhomology, and 3) targets exhibit variable but reproducible diversity with respect to the number and relative frequency of the mutational outcomes to which they give rise. From these data, we trained a model that uses local sequence context to predict the distribution of mutational outcomes. Exploiting the bias of NHEJ outcomes towards microhomology mediated events, we demonstrate the programming of deletion patterns by introducing microhomology to specific locations in the vicinity of the DSB site. We anticipate that our results will inform investigations of DSB repair mechanisms as well as the design of CRISPR/Cas9 experiments for diverse applications including genome-wide screens, gene therapy, lineage tracing and molecular recording.

genomics↗

bioRxiv2018

Multi-scale deep tensor factorization learns a latent representation of the human epigenome

The human epigenome has been experimentally characterized by measurements of protein binding, chromatin acessibility, methylation, and histone modification in hundreds of cell types. The result is a huge compendium of data, consisting of thousands of measurements for every basepair in the human genome. These data are difficult to make sense of, not only for humans, but also for computational methods that aim to detect genes and other functional elements, predict gene expression, characterize polymorphisms, etc. To address this challenge, we propose a deep neural network tensor factorization method, Avocado, that compresses epigenomic data into a dense, information-rich representation of the human genome. We use data from the Roadmap Epigenomics Consortium to demonstrate that this learned representation of the genome is broadly useful: first, by imputing epigenomic data more accurately than previous methods, and second, by showing that machine learning models that exploit this representation outperform those trained directly on epigenomic data on a variety of genomics tasks. These tasks include predicting gene expression, promoter-enhancer interactions, replication timing, and an element of 3D chromatin architecture. Our findings suggest the broad utility of Avocados learned latent representation for computational genomics and epigenomics.

bioinformatics↗

bioRxiv2017

Nucleotide sequence and DNaseI sensitivity are predictive of 3D chromatin architecture

Recently, Hi-C has been used to probe the 3D chromatin architecture of multiple organisms and cell types. The resulting collections of pairwise contacts across the genome have connected chromatin architecture to many cellular phenomena, including replication timing and gene regulation. However, high resolution (10 kb or finer) contact maps remain scarce due to the expense and time required for collection. A computational method for predicting pairwise contacts without the need to run a Hi-C experiment would be invaluable in understanding the role that 3D chromatin architecture plays in genome biology. We describe Rambutan, a deep convolutional neural network that predicts Hi-C contacts at 1 kb resolution using nucleotide sequence and DNaseI assay signal as inputs. Specifically, Rambutan identifies locus pairs that engage in high confidence contacts according to Fit-Hi-C, a previously described method for assigning statistical confidence estimates to Hi-C contacts. We first demonstrate Rambutans performance across chromosomes at 1 kb resolution in the GM12878 cell line. Subsequently, we measure Rambutans performance across six cell types. In this setting, the model achieves an area under the receiver operating characteristic curve between 0.7662 and 0.8246 and an area under the precision-recall curve between 0.3737 and 0.9008. We further demonstrate that the predicted contacts exhibit expected trends relative to histone modification ChlP-seq data, replication timing measurements, and annotations of functional elements such as promoters and enhancers. Finally, we predict Hi-C contacts for 53 human cell types and show that the predictions cluster by cellular function. [NOTE: After our original submission we discovered an error in our calling of statistically significant contacts. Briefly, when calculating the prior probability of a contact, we used the number of contacts at a certain genomic distance in a chromosome but divided by the total number of bins in the full genome. While we investigate what impact this had on our results, we ask that readers treat this manuscript skeptically.]

bioinformatics↗