dsMTL: a computational framework for privacy-preserving, distributed multi-task machine learning

Abstract

Motivation In multi-cohort machine learning studies, it is critical to differentiate between effects that are reproducible across cohorts and those that are cohort-specific. Multi-task learning (MTL) is a machine learning approach that facilitates this differentiation through the simultaneous learning of prediction tasks across cohorts. Since multi-cohort data can often not be combined into a single storage solution, there would be the substantial utility of an MTL application for geographically distributed data sources. Results Here, we describe the development of “dsMTL”, a computational framework for privacy-preserving, distributed multi-task machine learning that includes three supervised and one unsupervised algorithms. First, we derive the theoretical properties of these methods and the relevant machine learning workflows to ensure the validity of the software implementation. Second, we implement dsMTL as a library for the R programming language, building on the DataSHIELD platform that supports the federated analysis of sensitive individual-level data. Third, we demonstrate the applicability of dsMTL for comorbidity modeling in distributed data. We show that comorbidity modeling using dsMTL outperformed conventional, federated machine learning, as well as the aggregation of multiple models built on the distributed datasets individually. The application of dsMTL was computationally efficient and highly scalable when applied to moderate-size (n < 500), real expression data given the actual network latency. Availability dsMTL is freely available at https://github.com/transbioZI/dsMTLBase (server-side package) and https://github.com/transbioZI/dsMTLClient (client-side package). Supplementary information Supplementary data are available at Bioinformatics online.

Advancing translational research in neuroscience through multi-task learning

“…And the underlying true model is expected to be identified by considering the task-specific variations. This method has been used for the integration of multi-cohort gene expression datasets to identify expression signatures in brain samples from donors with schizophrenia (27). This study illustrated that MTL models showed higher reproducibility to unseen data cohorts than conventional machine learning methods and may thus be of particular use for the identification of reproducible biomarker signatures across studies.…”

“…This review aims to highlight MTL's utility for multi-modal data analysis. In biomedical applications, it is common to analyze the integration of heterogeneous but related data modalities, e.g., predictions at different time points during illness progression (26), case-control classification in different cohorts (27), or response prediction of multiple drugs (12). In psychiatric research, MTL has already been used to integrate schizophrenia markers from multiple cohorts (27) as well as measures of cognitive functioning and structural neuroimaging (28).…”

“…In biomedical applications, it is common to analyze the integration of heterogeneous but related data modalities, e.g., predictions at different time points during illness progression (26), case-control classification in different cohorts (27), or response prediction of multiple drugs (12). In psychiatric research, MTL has already been used to integrate schizophrenia markers from multiple cohorts (27) as well as measures of cognitive functioning and structural neuroimaging (28). By presenting MTL from different perspectives in the following sections, we hope to encourage the reader to identify useful applications in translational and neurogenetic research, such as the genetic prediction of illness risk and neural function (i.e., for identifying a genetic risk profile that shows neural effects) or the simultaneous prediction of diagnostic status and treatment response (i.e., to identify illness-associated biological patterns that are also responsive to treatment).…”

“…In translational psychiatry, MTL is a promising approach because it allows the integrative analysis of clinically- and biologically relevant data modalities. Specifically, in multi-modal data analysis, MTL can differentiate information shared between and specific for different modalities ( 29 ). Unlike machine learning, which estimates a parameter for a predictor (e.g., the expression value of a gene) within a data modality, MTL estimates the parameters of biologically related predictors (e.g., the expression value of a gene and the DNA methylation at the methylation sites in chromosomal proximity to the same gene) across modalities simultaneously through regularization.…”

“…For example, one study integrated COVID-19 patients’ information across multiple hospitals for training a prediction model using a variant of a federated algorithm ( 60 ). Recently, a user-friendly federated MTL software supporting multi-modal biomedical analysis has been developed ( 29 ).…”