Deep transfer learning provides aParetoimprovement for multi-ancestral clinico-genomic prediction of diseases

Abstract

Accurate predisposition assessment is essential for the prevention and early detection of diseases. Polygenic scores and machine learning models have been developed for disease prediction based on genetic variants and other risk factors. However, over 80% of genomic data were acquired from individuals of European descent. Other ethnic groups comprise the vast majority of the world population and have a severe data disadvantage. Due to the lack of suitable training data, clinico-genomic risk prediction is less accurate for the non-European population. Here we employ a transfer learning strategy to improve the clinico-genomic prediction of disease occurrence for data-disadvantaged populations. Our multiethnic machine learning experiments on real and synthetic datasets show that transfer learning can significantly improve disease prediction accuracy for data-disadvantaged populations. Under the transfer learning scheme, the prediction accuracy for the data-disadvantaged populations can be improved without compromising the prediction accuracy for other populations. Therefore, transfer learning provides a Pareto improvement toward equitable machine learning for genomic medicine.

Optimizing clinico-genomic disease prediction across ancestries: a machine learning strategy with Pareto improvement

“…The synthetic datasets are available from https:// figsh are. com/ artic les/ media/ TLGP_ GM/ 25377 532 [107]. The source code is available from https:// github.…”

Addressing the Challenge of Biomedical Data Inequality: An Artificial Intelligence Perspective

“…The current prevalent machine learning scheme for multiethnic data, the mixture learning scheme, and its main alternative, the independent learning scheme, have major obstacles in training optimal machine learning models for data-disadvantaged subpopulations (19,(93)(94)(95). The two Multiethnic machine learning schemes.…”

“…In transfer learning, a machine learning model trained on a data-rich subpopulation (source domain) can aid in training a model for a data-disadvantaged subpopulation (target domain) without affecting its own prediction accuracy. Thus, transfer learning provides a Pareto improvement (112) for multiethnic machine learning (95). Pareto improvement is a generally desired scenario in which some parties are better off without negatively impacting other parties in the system.…”

“…Machine learning experiments on synthetic data show that data inequality and subpopulation shift are the key factors underlying model performance disparities (19,95). Currently, these challenges in multiethnic machine learning are being addressed on two fronts: data collection and algorithmic intervention (Figure 6).…”