Accumulation processes, where many potentially coupled features are acquired over time, occur throughout the sciences from evolutionary biology to disease progression, and particularly in the study of cancer progression. Existing methods for learning the dynamics of such systems typically assume limited (often pairwise) relationships between feature subsets, cross-sectional or untimed observations, small feature sets, or discrete orderings of events. Here we introduce HyperTraPS-CT (Hypercubic Transition Path Sampling in Continuous Time) to compute posterior distributions on continuous-time dynamics of many, arbitrarily coupled, traits in unrestricted state spaces, accounting for uncertainty in observations and their timings. We demonstrate the capacity of HyperTraPS-CT to deal with cross-sectional, longitudinal, and phylogenetic data, which may have no, uncertain, or precisely specified sampling times. HyperTraPS-CT allows positive and negative interactions between arbitrary subsets of features (not limited to pairwise interactions), supporting Bayesian and maximum-likelihood inference approaches to identify these interactions, consequent pathways, and predictions of future and unobserved features. We also introduce a range of visualisations for the inferred outputs of these processes and demonstrate model selection and regularisation for feature interactions. We apply this approach to case studies on the accumulation of mutations in cancer progression and the acquisition of anti-microbial resistance genes in tuberculosis, demonstrating its flexibility and capacity to produce predictions aligned with applied priorities.

Can we understand and predict the evolutionary pathways by which bacteria acquire multi-drug resistance (MDR)? These questions have substantial potential impact in basic biology and in applied approaches to address the global health challenge of antimicrobial resistance (AMR). In this minireview, we discuss how a class of machine-learning approaches called evolutionary accumulation modeling (EvAM) may help reveal these dynamics using genetic and/or phenotypic AMR data sets, without requiring longitudinal sampling. These approaches are well-established in cancer progression and evolutionary biology but currently less used in AMR research. We discuss how EvAM can learn the evolutionary pathways by which drug resistances and other AMR features (for example, mutations driving these resistances) are acquired as pathogens evolve, predict next evolutionary steps, identify influences between AMR features, and explore differences in MDR evolution between regions, demographics, and more. We demonstrate a case study from the literature on MDR evolution in Mycobacterium tuberculosis and discuss the strengths and weaknesses of these approaches, providing links to some approaches for implementation.

A common problem in medical applications is predicting the time until an event of interest such as the onset of a disease, time to tumor recurrence, and time to mortality. Traditionally, classical survival analysis techniques have been used to address this problem. However, these techniques are of limited usage when considering nonlinear and interaction effects among biomarkers, and high profiling survival datasets. Although supervised machine learning techniques have shown some advantages over standard statistical methods in handling high-dimensional datasets, their application to survival analysis is at best limited. A major reason behind this is the difficulty in processing censored data, which is a common component of survival analysis. In this paper, we have transformed the time-to-event prediction problem into a semi-supervised regression problem in which we use a self-training wrapper approach with random survival forests as the base learner. In this approach, censored observations are introduced as partially labeled observations since their predicted time (target value) should exceed the censoring time. First, the algorithm builds a base model over the observed instances and then augments them iteratively with highly confident predictions over the censored set, using a smart stopping criterion based on the censoring time. The proposed approach has been evaluated and compared on fifteen real-world survival analysis datasets, including clinical and high-dimensional data. The ability of our proposed approach to integrate partial supervision information within a semi-supervised learning strategy has enabled it to achieve competitive performance compared to baseline models, particularly in the case of a high-dimensional regime.

Accumulation modelling uses machine learning to discover the dynamics by which systems acquire discrete features over time. Many systems of biomedical interest show such dynamics: from bacteria acquiring resistances to sets of drugs, to patients acquiring symptoms during the course of progressive disease. Existing approaches for accumulation modelling are typically limited either in the number of features they consider or their ability to characterise interactions between these features – a limitation for the large-scale genetic and/or phenotypic datasets often found in modern biomedical applications. Here, we demonstrate how clustering can make such large-scale datasets tractable for powerful accumulation modelling approaches. Clustering resolves issues of sparsity and high dimensionality in datasets, but complicates the intepretation of the inferred dynamics, especially if observations are not independent. Focussing on hypercubic hidden Markov models (HyperHMM), we introduce several approaches for interpreting, estimating, and bounding the results of the dynamics in these cases and discuss how biomedical insight could be gained from such analyses. We demonstrate this ‘Cluster-based HyperHMM’ (CHyperHMM) pipeline for synthetic data, clinical data on disease progression in severe malaria, and genomic data for anti-microbial resistance evolution in Klebsiella pneumoniae , reflecting two global health threats.

Mitochondria and plastids have both dramatically reduced their genomes since the endosymbiotic events that created them. The similarities and differences in the evolution of the two organelle genome types have been the target of discussion and investigation for decades. Ongoing work has suggested that similar mechanisms may modulate the reductive evolution of the two organelles in a given species, but quantitative data and statistical analyses exploring this picture remain limited outside of some specific cases like parasitism. Here, we use cross-eukaryote organelle genome data to explore evidence for coevolution of mitochondrial and plastid genome reduction. Controlling for differences between clades and pseudoreplication due to relatedness, we find that extents of mtDNA and ptDNA gene retention are related to each other across taxa, in a generally positive correlation that appears to differ quantitatively across eukaryotes, for example, between algal and nonalgal species. We find limited evidence for coevolution of specific mtDNA and ptDNA gene pairs, suggesting that the similarities between the two organelle types may be due mainly to independent responses to consistent evolutionary drivers.

High-dimensional data have attracted considerable interest from researchers, especially in the area of variable selection. However, when dealing with time-to-event data in survival analysis, where censoring is a key consideration, progress in addressing this complex problem has remained somewhat limited. Moreover, in microarray research, it is common to identify groupings of genes involved in the same biological pathways. These gene groupings frequently collaborate and operate as a unified entity. Therefore, this study is motivated to adopt the idea of a penalized semi-parametric Bayesian Cox (PSBC) model through elastic-net and group lasso penalty functions (PSBC-EN and PSBC-GL) to incorporate the grouping structure of the covariates (genes) and optimally perform variable selection. The proposed methods assign a beta process prior to the cumulative baseline hazard function (PSBC-EN-B and PSBC-GL-B), instead of the gamma process prior used in existing methods (PSBC-EN-G and PSBC-GL-G). Three real-life datasets and simulation scenarios were considered to compare and validate the efficiency of the modified methods with existing techniques, using Bayesian information criteria (BIC). The results of the simulated studies provided empirical evidence that the proposed methods performed better than the existing methods across a wide range of data scenarios. Similarly, the results of the real-life study showed that the proposed methods revealed a substantial improvement over the existing techniques in terms of feature selection and grouping behavior.