arXiv subjects

Thomas P. Quinn

Publications and source records attributed to Thomas P. Quinn.

7 recordsLinked to original sources

arXiv2023

Bugs as Features (Part I): Concepts and Foundations for the Compositional Data Analysis of the Microbiome-Gut-Brain Axis

There has been a growing acknowledgement of the involvement of the gut microbiome - the collection of microbes that reside in our gut - in regulating our mood and behaviour. This phenomenon is referred to as the microbiome-gut-brain axis. While our techniques to measure the presence and abundance of these microbes have been steadily improving, the analysis of microbiome data is non-trivial. Here, we present a perspective on the concepts and foundations of data analysis and interpretation of microbiome experiments with a focus on the microbiome-gut-brain axis domain. We give an overview of foundational considerations prior to commencing analysis alongside the core microbiome analysis approaches of alpha diversity, beta diversity, differential feature abundance and functional inference. We emphasize the compositional data analysis (CoDA) paradigm. Further, this perspective features an extensive and heavily annotated microbiome analysis in R in the supplementary materials, as a resource for new and experienced bioinformaticians alike.

q-bio.GN↗

arXiv2023

Knowledge-based Integration of Multi-Omic Datasets with Anansi: Annotation-based Analysis of Specific Interactions

Motivation: Studies including more than one type of 'omics data sets are becoming more prevalent. Integrating these data sets can be a way to solidify findings and even to make new discoveries. However, integrating multi-omics data sets is challenging. Typically, data sets are integrated by performing an all-vs-all correlation analysis, where each feature of the first data set is correlated to each feature of the second data set. However, all-vs-all association testing produces unstructured results that are hard to interpret, and involves potentially unnecessary hypothesis testing that reduces statistical power due to false discovery rate (FDR) adjustment. Implementation: Here, we present the anansi framework, and accompanying R package, as a way to improve upon all-vs-all association analysis. We take a knowledge-based approach where external databases like KEGG are used to constrain the all-vs-all association hypothesis space, only considering pairwise associations that are a priori known to occur. This produces structured results that are easier to interpret, and increases statistical power by skipping unnecessary hypothesis tests. In this paper, we present the anansi framework and demonstrate its application to learn metabolite-function interactions in the context of host-microbe interactions. We further extend our framework beyond pairwise association testing to differential association testing, and show how anansi can be used to identify associations that differ in strength or degree based on sample covariates such as case/control status. Availability: https://github.com/thomazbastiaanssen/anansi

q-bio.BM↗

arXiv2022

Data Augmentation for Compositional Data: Advancing Predictive Models of the Microbiome

Data augmentation plays a key role in modern machine learning pipelines. While numerous augmentation strategies have been studied in the context of computer vision and natural language processing, less is known for other data modalities. Our work extends the success of data augmentation to compositional data, i.e., simplex-valued data, which is of particular interest in the context of the human microbiome. Drawing on key principles from compositional data analysis, such as the Aitchison geometry of the simplex and subcompositions, we define novel augmentation strategies for this data modality. Incorporating our data augmentations into standard supervised learning pipelines results in consistent performance gains across a wide range of standard benchmark datasets. In particular, we set a new state-of-the-art for key disease prediction tasks including colorectal cancer, type 2 diabetes, and Crohn's disease. In addition, our data augmentations enable us to define a novel contrastive learning model, which improves on previous representation learning approaches for microbiome compositional data. Our code is available at https://github.com/cunningham-lab/AugCoDa.

stat.ML↗

arXiv2021

Stool Studies Don't Pass the Sniff Test: A Systematic Review of Human Gut Microbiome Research Suggests Widespread Misuse of Machine Learning

In the machine learning culture, an independent test set is required for proper model verification. Failures in model verification, including test set omission and test set leakage, make it impossible to know whether or not a trained model is fit for purpose. In this article, we present a systematic review and quantitative analysis of human gut microbiome classification studies, conducted to measure the frequency and impact of test set omission and test set leakage on area under the receiver operating curve (AUC) reporting. Among 102 articles included for analysis, we find that only 12% of studies report a bona fide test set AUC, meaning that the published AUCs for 88% of studies cannot be trusted at face value. Our findings cast serious doubt on the general validity of research claiming that the gut microbiome has high diagnostic or prognostic potential in human disease.

q-bio.GN↗

arXiv2020

The Three Ghosts of Medical AI: Can the Black-Box Present Deliver?

Our title alludes to the three Christmas ghosts encountered by Ebenezer Scrooge in \textit{A Christmas Carol}, who guide Ebenezer through the past, present, and future of Christmas holiday events. Similarly, our article will take readers through a journey of the past, present, and future of medical AI. In doing so, we focus on the crux of modern machine learning: the reliance on powerful but intrinsically opaque models. When applied to the healthcare domain, these models fail to meet the needs for transparency that their clinician and patient end-users require. We review the implications of this failure, and argue that opaque models (1) lack quality assurance, (2) fail to elicit trust, and (3) restrict physician-patient dialogue. We then discuss how upholding transparency in all aspects of model design and model validation can help ensure the reliability of medical AI.

cs.AI↗

arXiv2020

Trust and Medical AI: The challenges we face and the expertise needed to overcome them

Artificial intelligence (AI) is increasingly of tremendous interest in the medical field. However, failures of medical AI could have serious consequences for both clinical outcomes and the patient experience. These consequences could erode public trust in AI, which could in turn undermine trust in our healthcare institutions. This article makes two contributions. First, it describes the major conceptual, technical, and humanistic challenges in medical AI. Second, it proposes a solution that hinges on the education and accreditation of new expert groups who specialize in the development, verification, and operation of medical AI technologies. These groups will be required to maintain trust in our healthcare institutions.

cs.AI↗

arXiv2020

DeepCoDA: personalized interpretability for compositional health data

Interpretability allows the domain-expert to directly evaluate the model's relevance and reliability, a practice that offers assurance and builds trust. In the healthcare setting, interpretable models should implicate relevant biological mechanisms independent of technical factors like data pre-processing. We define personalized interpretability as a measure of sample-specific feature attribution, and view it as a minimum requirement for a precision health model to justify its conclusions. Some health data, especially those generated by high-throughput sequencing experiments, have nuances that compromise precision health models and their interpretation. These data are compositional, meaning that each feature is conditionally dependent on all other features. We propose the Deep Compositional Data Analysis (DeepCoDA) framework to extend precision health modelling to high-dimensional compositional data, and to provide personalized interpretability through patient-specific weights. Our architecture maintains state-of-the-art performance across 25 real-world data sets, all while producing interpretations that are both personalized and fully coherent for compositional data.

cs.LG↗