Molecular Heterogeneity in Large-Scale Biological Data: Techniques and Applications

Abstract

High-throughput sequencing technologies have evolved at a stellar pace for almost a decade and have greatly advanced our understanding of genome biology. In these sampling-based technologies, there is an important detail that is often overlooked in the analysis of the data and the design of the experiments, specifically that the sampled observations often do not give a representative picture of the underlying population. This has long been recognized as a problem in statistical ecology and in the broader statistics literature. In this review, we discuss the connections between these fields, methodological advances that parallel both the needs and opportunities of large-scale data analysis, and specific applications in modern biology. In the process we describe unique aspects of applying these approaches to sequencing technologies, including sequencing error, population and individual heterogeneity, and the design of experiments.

Bayesian Nonparametric Inference for "Species-sampling" Problems

“…In this paper, we present an overview of BNP inference for SSPs under the popular PYP prior. Motivated by the work of Deng et al [2019], we focus on SSPs corresponding to the aforementioned questions Q1, Q2 and Q3, respectively, which have been proved to be of practical interest also beyond biological sciences. Regarding Q1, we consider the estimation of coverage probabilities, which include the missing mass and the coverage probability of order r ≥ 1, namely the probability mass of species observed with frequency r in the sample.…”

“…In genomics data, they appear in relation to coverage depth, i.e. the average number of reads that are aligned to known reference bases [Deng et al, 2019]. Depending on the specific application, different levels of coverage might be required.…”

“…Biological sciences are arguably the field where SSPs have been most investigated over the past three decades, both in methods and applications. A comprehensive and up-to-date review is presented by Deng et al [2019], from which it emerges the critical role of SSPs for the understanding of genome biology. Although sequencing technologies have greatly advanced the understanding of genome biology, it is well known that observed samples are not perfectly representative of the molecular heterogeneity or species composition of the underlying DNA library, often providing a poor representation due to low-abundance molecules that are hard to sample.…”

“…Unless DNA libraries can be sequenced up to complete saturation, which is typically not convenient due to budget constraints, it is common to make use of the observed sample to infer the molecular heterogeneity of additional unobservable samples from the library, as well as of the library itself. Deng et al [2019] identified three main questions that a researcher may ask: Q1) what is the expected population frequency of a species with frequency r ≥ 1 in the sample?…”

A Bayesian Nonparametric Approach to Species Sampling Problems with Ordering

“…missing mass, discovery probabilities, unseen species with prevalences and coverages of prevalence. We refer to Deng et al (2019) and Balocchi et al (2022) for up-to-date reviews of SSPs, both in methods and applications, mostly in the field of biological sciences but also in machine learning, electrical engineering, theoretical computer science and information theory.…”