A Misclassification Network-Based Method for Comparative Genomic Analysis

TL;DR

GMNA presents an alignment-free framework that leverages misclassification patterns from AI classifiers to build a genome misclassification network, enabling scalable inference of geographic and mobility-driven structure in large genome ensembles. By defining indistinguishability between genome groups and using concepts such as LOCO training and soft misclassification, GMNA integrates AI with network analysis to uncover drivers of genomic variation beyond traditional accuracy metrics. The study demonstrates strong spatial dependencies in SARS-CoV-2 genomes, with edge weights correlating with travel-network centrality, and shows robustness across binary and multiclass settings as well as baseline k-mer NB and CNN/DNN variants. The framework offers a computationally efficient tool for large-scale comparative genomics with practical implications for understanding how human mobility shapes pathogen evolution and geographic clustering, supported by publicly available code and data resources.

Abstract

Classifying genome sequences based on metadata has been an active area of research in comparative genomics for decades with many important applications across the life sciences. Established methods for classifying genomes can be broadly grouped into sequence alignment-based and alignment-free models. Conventional alignment-based models rely on genome similarity measures calculated based on local sequence alignments or consistent ordering among sequences. However, such methods are computationally expensive when dealing with large ensembles of even moderately sized genomes. In contrast, alignment-free (AF) approaches measure genome similarity based on summary statistics in an unsupervised setting and are efficient enough to analyze large datasets. However, both alignment-based and AF methods typically assume fixed scoring rubrics that lack the flexibility to assign varying importance to different parts of the sequences based on prior knowledge. In this study, we integrate AI and network science approaches to develop a comparative genomic analysis framework that addresses these limitations. Our approach, termed the Genome Misclassification Network Analysis (GMNA), simultaneously leverages misclassified instances, a learned scoring rubric, and label information to classify genomes based on associated metadata and better understand potential drivers of misclassification. We evaluate the utility of the GMNA using Naive Bayes and convolutional neural network models, supplemented by additional experiments with transformer-based models, to construct SARS-CoV-2 sampling location classifiers using over 500,000 viral genome sequences and study the resulting network of misclassifications. We demonstrate the global health potential of the GMNA by leveraging the SARS-CoV-2 genome misclassification networks to investigate the role human mobility played in structuring geographic clustering of SARS-CoV-2.

Figures (9)

• Figure 1: Flowchart of the Genome Misclassification Network Analysis Framework.
• Figure 2: MCC Misclassification vs. Configuration Clustered by the Louvain Community Detection Algorithm: Our results show the SARS-CoV-2 genomes exhibit strong spatial dependencies. Genomes from neighboring regions are highly indistinguishable, and the significance of the findings is supported by the configuration model test.
• Figure 3: Network Visualization of Misclassification Results in Fig. \ref{['subfig:MCCNBtrue']}: In this genome misclassification network, each node represents a subset of genome sequences from a specific region, with edge weights representing the indistinguishability score, computed as the symmetrised empirical likelihood of misclassification between genomes from the two regions. The partition and coloring of the regions are based on community detection clustering results, with singular communities consisting of only one region colored in grey. This plot shows the top 20% edges with the greatest weights.
• Figure 4: Binary Naive Bayes Misclassification: The spatial dependencies among COVID genomes are also evident under the binary classification setting. Each subplot corresponds to a different centroid country, and a star misclassification network is constructed based on the Naive Bayes binary classification results relative to the centroid country. In (c) and (d), we are also showing the Europe close-ups of the misclassification networks on the right.
• Figure 5: Binary NB Misclassification Network vs. Configuration: The significance of the spatial dependencies is supported by the configuration model test. The misclassification network constructed under the binary scheme is a star network, relative to the centroid England. The redness of the other regions are determined by its indistinguishability score with resepct to genomes from England.