Data Heterogeneity: The Enzyme to Catalyze Translational Bioinformatics?

Eli M Cahan,Purvesh Khatri

doi:10.2196/18044

Abstract

Up to 95% of novel interventions demonstrating significant effects at the bench fail to translate to the bedside. In recent years, the windfalls of “big data” have afforded investigators more substrate for research than ever before. However, issues with translation have persisted: although countless biomarkers for diagnostic and therapeutic targeting have been proposed, few of these generalize effectively. We assert that inadequate heterogeneity in datasets used for discovery and validation causes their nonrepresentativeness of the diversity observed in real-world patient populations. This nonrepresentativeness is contrasted with advantages rendered by the solicitation and utilization of data heterogeneity for multisystemic disease modeling. Accordingly, we propose the potential benefits of models premised on heterogeneity to promote the Institute for Healthcare Improvement’s Triple Aim. In an era of personalized medicine, these models can confer higher quality clinical care for individuals, increased access to effective care across all populations, and lower costs for the health care system.

Highlights

Philosopher Karl Popper commented in 1934 that “non-reproducible single occurrences are of no significance to science” [1]
85 years since this statement was made, science remains inundated with nonreproducible single occurrences
Up to 89% of studies demonstrating significant preclinical effects of novel molecules are nonreplicable [4], and the translation failure rate of novel interventions demonstrating significant effects preclinically that are never approved for clinical use reaches up to 95% [5]

Summary

Background

Philosopher Karl Popper commented in 1934 that “non-reproducible single occurrences are of no significance to science” [1]. There is already evidence for the effectiveness of translational bioinformatics premised on heterogeneity for conditions previously plagued by generalization failures, such as in the derivation of host response–based gene panels to predict sepsis and tuberculosis These panels have outperformed all precedents developed without accounting for heterogeneity (including those using the most sophisticated machine-learning techniques); have been validated across time points, disease severity cohorts, and comorbidities; and have been generalizable across multiple continents [7,8,9]. It has been noted that increasing the number of datasets included in GWAS samples, controlling for sample size, markedly improves the predictive power of the obtained gene panels to a much greater extent than expanding the sample size alone [15] This model incorporates “dark matter” contributing to “missing heritability,” permitting the parsimonious identification of key biological pathways in spite of environmental differences between patient cohorts [16,38]. In this way, channeling the “wisdom of crowds” refers to wisdom pulled by collaboration between investigators and to the wisdom pushed by the comprehensiveness of study populations

Conclusion

Findings

21. Khatri Lab