Predicting correlated outcomes from molecular data

A decision-analytical perspective on incorporating multiple outcomes in the production of clinical prediction models: defining a taxonomy of risk estimands

BACKGROUND

Clinical prediction models (CPMs) estimate an individual's risk of current or future outcome events, using information available about the individual at the time of prediction. While most CPMs are developed to predict a single outcome event, many clinical decisions require considering the risks of multiple outcome events. For example, decision-making for anticoagulation therapy involves assessing an individual's risks of both blood clot and bleeding, while decision-making around interventions for multimorbidity prevention requires an understanding of the risks of developing multiple long-term conditions. However, determining when and how to incorporate multiple outcomes into CPMs remains challenging. This article aims to raise awareness of multiple outcome prediction and present clinical examples where such prediction is essential to help inform individual decision-making.

MAIN TEXT

A range of analytical methods are available to develop multiple-outcome CPMs, but there are frequent malapropisms and heterogeneity in terminology across this literature, making it difficult to identify/compare possible methods. Selecting the appropriate method should depend on the intended risk estimand-the type of predicted risks that we wish the CPM to estimate-but this is often not defined or reported. Using clinical examples and a decision-analytical perspective, we present a taxonomy of risk estimands to frame different clinical contexts requiring multiple-outcome CPMs. We outline four levels of risk estimands: (i) single-outcome risk, (ii) competing-outcome risk, (iii) composite-outcome risk, and (iv) risk of multiple outcome combinations. We demonstrate how a decision-analytical and utility-theory lens can help define the risk estimand for a given clinical scenario, based on the model's intended use.

CONCLUSIONS

Clearly defining and reporting the risk estimand is essential for all prediction model studies. A decision-analytical framework aids in selecting the most appropriate estimand for a given prediction task and in determining when and how to incorporate multiple outcomes into CPM development.

Penalized regression with multiple sources of prior effects

MOTIVATION

In many high-dimensional prediction or classification tasks, complementary data on the features are available, e.g. prior biological knowledge on (epi)genetic markers. Here we consider tasks with numerical prior information that provide an insight into the importance (weight) and the direction (sign) of the feature effects, e.g. regression coefficients from previous studies.

RESULTS

We propose an approach for integrating multiple sources of such prior information into penalized regression. If suitable co-data are available, this improves the predictive performance, as shown by simulation and application.

AVAILABILITY AND IMPLEMENTATION

The proposed method is implemented in the R package transreg (https://github.com/lcsb-bds/transreg, https://cran.r-project.org/package=transreg).

Isoform-level transcriptome-wide association uncovers genetic risk mechanisms for neuropsychiatric disorders in the human brain

Methods integrating genetics with transcriptomic reference panels prioritize risk genes and mechanisms at only a fraction of trait-associated genetic loci, due in part to an overreliance on total gene expression as a molecular outcome measure. This challenge is particularly relevant for the brain, in which extensive splicing generates multiple distinct transcript-isoforms per gene. Due to complex correlation structures, isoform-level modeling from cis-window variants requires methodological innovation. Here we introduce isoTWAS, a multivariate, stepwise framework integrating genetics, isoform-level expression and phenotypic associations. Compared to gene-level methods, isoTWAS improves both isoform and gene expression prediction, yielding more testable genes, and increased power for discovery of trait associations within genome-wide association study loci across 15 neuropsychiatric traits. We illustrate multiple isoTWAS associations undetectable at the gene-level, prioritizing isoforms of AKT3, CUL3 and HSPD1 in schizophrenia and PCLO with multiple disorders. Results highlight the importance of incorporating isoform-level resolution within integrative approaches to increase discovery of trait associations, especially for brain-relevant traits.

Predicting dichotomised outcomes from high-dimensional data in biomedicine

In many biomedical applications, we are more interested in the predicted probability that a numerical outcome is above a threshold than in the predicted value of the outcome. For example, it might be known that antibody levels above a certain threshold provide immunity against a disease, or a threshold for a disease severity score might reflect conversion from the presymptomatic to the symptomatic disease stage. Accordingly, biomedical researchers often convert numerical to binary outcomes (loss of information) to conduct logistic regression (probabilistic interpretation). We address this bad statistical practice by modelling the binary outcome with logistic regression, modelling the numerical outcome with linear regression, transforming the predicted values from linear regression to predicted probabilities, and combining the predicted probabilities from logistic and linear regression. Analysing high-dimensional simulated and experimental data, namely clinical data for predicting cognitive impairment, we obtain significantly improved predictions of dichotomised outcomes. Thus, the proposed approach effectively combines binary with numerical outcomes to improve binary classification in high-dimensional settings. An implementation is available in the R package cornet on GitHub (https://github.com/rauschenberger/cornet) and CRAN (https://CRAN.R-project.org/package=cornet).

MAK: a machine learning framework improved genomic prediction via multi-target ensemble regressor chains and automatic selection of assistant traits

Incorporating the genotypic and phenotypic of the correlated traits into the multi-trait model can significantly improve the prediction accuracy of the target trait in animal and plant breeding, as well as human genetics. However, in most cases, the phenotypic information of the correlated and target trait of the individual to be evaluated was null simultaneously, particularly for the newborn. Therefore, we propose a machine learning framework, MAK, to improve the prediction accuracy of the target trait by constructing the multi-target ensemble regression chains and selecting the assistant trait automatically, which predicted the genomic estimated breeding values of the target trait using genotypic information only. The prediction ability of MAK was significantly more robust than the genomic best linear unbiased prediction, BayesB, BayesRR and the multi trait Bayesian method in the four real animal and plant datasets, and the computational efficiency of MAK was roughly 100 times faster than BayesB and BayesRR.

Differential Phospho-Signatures in Blood Cells Identify LRRK2 G2019S Carriers in Parkinson's Disease

Background

The clinicopathological phenotype of G2019S LRRK2‐associated Parkinson's disease (L2PD) is similar to idiopathic Parkinson's disease (iPD), and G2019S LRRK2 nonmanifesting carriers (L2NMCs) are at increased risk for development of PD. With various therapeutic strategies in the clinical and preclinical pipeline, there is an urgent need to identify biomarkers that can aid early diagnosis and patient enrichment for ongoing and future LRRK2‐targeted trials.

Objective

The objective of this work was to investigate differential protein and phospho‐protein changes related to G2019S mutant LRRK2 in peripheral blood mononuclear cells from G2019S L2PD patients and G2019S L2NMCs, identify specific phospho‐protein changes associated with the G2019S mutation and with disease status, and compare findings with patients with iPD.

Methods

We performed an unbiased phospho‐proteomic study by isobaric label–based mass spectrometry using peripheral blood mononuclear cell group pools from a LRRK2 cohort from Spain encompassing patients with G2019S L2PD (n = 20), G2019S L2NMCs (n = 20), healthy control subjects (n = 30), patients with iPD (n = 15), patients with R1441G L2PD (n = 5), and R1441G L2NMCs (n = 3) (total N = 93).

Results

Comparing G2019S carriers with healthy controls, we identified phospho‐protein changes associated with the G2019S mutation. Moreover, we uncovered a specific G2019S phospho‐signature that changes with disease status and can discriminate patients with G2019S L2PD, G2019S L2NMCs, and healthy controls. Although patients with iPD showed a differential phospho‐proteomic profile, biological enrichment analyses revealed similar changes in deregulated pathways across the three groups.

Conclusions

We found a differential phospho‐signature associated with LRRK2 G2019S for which, consistent with disease status, the phospho‐profile from PD at‐risk G2019S L2NMCs was more similar to healthy controls than patients with G2019S L2PD with the manifested disease. © 2022 The Authors. Movement Disorders published by Wiley Periodicals LLC on behalf of International Parkinson and Movement Disorder Society