NCBoost classifies pathogenic non-coding variants in Mendelian diseases through supervised learning on purifying selection signals in humans

Deep learning models have shown great promise in predicting genome-wide regulatory effects from DNA sequence, but their informativeness for human complex diseases and traits is not fully understood. Here, we evaluate the disease informativeness of two types of deep learning annotations: (1) variant-level annotations (based on the reference allele), assessing whether they are more informative for complex disease than the underlying experimental data used to train the predictive models; and (2) allelic-effect annotations (absolute value of the predicted difference between reference and variant alleles), which have been a major focus of recent work. In each case, we primarily consider annotations constructed using two previously trained deep learning models, DeepSEA and Basenji. We apply stratified LD score regression (S-LDSC) to 41 independent diseases and complex traits (average N =320K) to evaluate each annotation's informativeness for disease heritability conditional on a broad set of coding, conserved, regulatory and LDrelated annotations from the baseline-LD model and other sources; as a secondary metric, we also evaluate the accuracy of models that incorporate deep learning annotations in predicting disease-associated or fine-mapped SNPs. We aggregated annotations across all tissues (resp. blood cell types or brain tissues) in metaanalyses across all 41 traits (resp. 11 blood-related traits or 8 brain-related traits). Variant-level annotations, despite being highly enriched for disease heritability, produced no conditionally significant results in meta-analyses across all 41 traits or 11 blood-related traits, but brain-specific DeepSEA-H3K4me3 and Basenji-H3K27ac annotations were conditionally significant in meta-analyses across 8 brain-related traits; a sequence motif analysis suggests that these annotations could be capturing unique information about nucleosome occupancy. Allelic-effect annotations were also highly enriched for disease heritability, and produced conditionally significant results for Basenji-H3K4me3 in meta-analyses across all 41 traits and brain-specific Basenji-H3K4me3 in meta-analyses across 8 brain-related traits. We conclude that deep learning models are informative for disease, but their informativeness cannot be inferred from metrics based on their accuracy in predicting regulatory annotations.

show abstract

“…To optimize classification performance, we selected XGBoost parameter settings to minimize overfitting, as in ref. 58 .…”

Section: Classification Of Disease-associated or Fine-mapped Snpsmentioning

confidence: 99%

Evaluating the informativeness of deep learning annotations for human complex diseases

Dey

Geijn

Kim

et al. 2019

Preprint

View full text Add to dashboard Cite

show abstract

“…We note three key differences between AnnotBoost and previous approaches that utilized gradient boosting to identify pathogenic missense 7 and non-coding variants 9,10 . First, AnnotBoost uses a pathogenicity score as the only input and does not use disease data (e.g.…”

Section: Discussionmentioning

confidence: 98%

“…Second, AnnotBoost produces genome-wide scores, even when some SNPs are unscored by the input pathogenicity score. Third, AnnotBoost leverages 75 diverse features from the baseline-LD model 26,27 , significantly more than previous approaches 7, 9,10 . Indeed, we determined that AnnotBoost produces strong signals even when conditioned on those approaches.…”

Section: Discussionmentioning

confidence: 99%

Improving the informativeness of Mendelian disease-derived pathogenicity scores for common disease

Kim

Dey

Weissbrod

et al. 2020

Preprint

View full text Add to dashboard Cite

Despite considerable progress on pathogenicity scores prioritizing both coding and non-coding variants for Mendelian disease, little is known about the utility of these pathogenicity scores for common disease. Here, we sought to assess the informativeness of Mendelian disease-derived pathogenicity scores for common disease, and to improve upon existing scores. We first applied stratified LD score regression to assess the informativeness of annotations defined by top variants from published Mendelian disease-derived pathogenicity scores across 41 independent common diseases and complex traits (average N = 320K). Several of the resulting annotations were informative for common disease, even after conditioning on a broad set of coding, conserved, regulatory and LD-related annotations from the baseline-LD model. We then improved upon the published pathogenicity scores by developing AnnotBoost, a gradient boosting-based framework to impute and denoise pathogenicity scores using functional annotations from the baseline-LD model. AnnotBoost substantially increased the informativeness for common disease of both previously uninformative and previously informative pathogenicity scores, implying pervasive variant-level overlap between Mendelian disease and common disease. The boosted scores also produced significant improvements in heritability model fit and in classifying disease-associated, fine-mapped SNPs. Our boosted scores have high potential to improve candidate gene discovery and fine-mapping for common disease.

show abstract

“…As recently pointed out in [28], pathogenic scores predicted by several state-of-the-art methods are biased towards some specific regulatory region types. Indeed also with Mendelian and GWAS data the positive set of variants is located in different functional non-coding regions (like 5'UTR, 3'UTR or Promoter) and is not evenly distributed over them.…”

Section: Assessment Of the Effect On Prediction Performance Of The Vamentioning

confidence: 97%

“…The first one (GWAVA) applied a modified random forest [26], where its decision trees are trained on artificially balanced data, thus reducing the imbalance of the data [27]. A second one (NCBoost) used gradient tree boosting learning machines with partially balanced data, achieving very competitive results in the prioritization of pathogenic Mendelian variants, even if the comparison with the other state-of-the-art methods have been performed without retraining them, but using only their pre-computed scores [28]. The unbalancing issue has been fully addressed by ReMM [29] and hyperSMURF [30], through the application of subsampling techniques to the "negative" neutral variants, and oversampling algorithms to the set of "positive" pathogenic variants.…”

Section: Introductionmentioning

confidence: 99%

parSMURF, a High Performance Computing tool for the genome-wide detection of pathogenic variants

Petrini

Mesiti

Schubach

et al. 2020

Preprint

View full text Add to dashboard Cite

Several prediction problems in Computational Biology and Genomic Medicine are characterized by both big data as well as a high imbalance between examples to be learned, whereby positive examples can represent a tiny minority with respect to negative examples. For instance, deleterious or pathogenic variants are overwhelmed by the sea of neutral variants in the non-coding regions of the genome: as a consequence the prediction of deleterious variants is a very challenging highly imbalanced classification problem, and classical prediction tools fail to detect the rare pathogenic examples among the huge amount of neutral variants or undergo severe restrictions in managing big genomic data. To overcome these limitations we propose parSMURF, a method that adopts a hyper-ensemble approach and oversampling and undersampling techniques to deal with imbalanced data, and parallel computational techniques to both manage big genomic data and significantly speed-up the computation. The synergy between Bayesian optimization techniques and the parallel nature of parSMURF enables efficient and user-friendly automatic tuning of the hyper-parameters of the algorithm, and allows specific learning problems in Genomic Medicine to be easily fit. Moreover, by using MPI parallel and machine learning ensemble techniques, parSMURF can manage big data by partitioning them across the nodes of a High Performance Computing cluster. Results with synthetic data and with single nucleotide variants associated with Mendelian diseases and with GWAS hits in the non-coding regions of the human genome, involving millions of examples, show that parSMURF achieves state-of-the-art results and a speed-up of 80× with respect to the sequential version. In conclusion parSMURF is a parallel machine learning tool that can be trained to learn different genomic problems, and its multiple levels of parallelization and its high scalability allow us to efficiently fit problems characterized by big and imbalanced genomic data. Availability and Implementation: The C++ OpenMP multi-core version tailored to a single workstation and the C++ MPI/OpenMP hybrid multi-core and multi-node parSMURF version tailored to a High Performance Computing cluster are both available from github: https://github.com/AnacletoLAB/parSMURF

show abstract

NCBoost classifies pathogenic non-coding variants in Mendelian diseases through supervised learning on purifying selection signals in humans

Cited by 57 publications

References 71 publications

Evaluating the informativeness of deep learning annotations for human complex diseases

Evaluating the informativeness of deep learning annotations for human complex diseases

Improving the informativeness of Mendelian disease-derived pathogenicity scores for common disease

parSMURF, a High Performance Computing tool for the genome-wide detection of pathogenic variants

Contact Info

Product

Resources

About