Performance and Information Leakage in Splitfed Learning and Multi-Head Split Learning in Healthcare Data and Beyond.

Machine learning (ML) in healthcare data analytics is attracting much attention because of the unprecedented power of ML to extract knowledge that improves the decision-making process. At the same time, laws and ethics codes drafted by countries to govern healthcare data are becoming stringent. Although healthcare practitioners are struggling with an enforced governance framework, we see the emergence of distributed learning-based frameworks disrupting traditional-ML-model development. Splitfed learning (SFL) is one of the recent developments in distributed machine learning that empowers healthcare practitioners to preserve the privacy of input data and enables them to train ML models. However, SFL has some extra communication and computation overheads at the client side due to the requirement of client-side model synchronization. For a resource-constrained client side (hospitals with limited computational powers), removing such conditions is required to gain efficiency in the learning. In this regard, this paper studies SFL without client-side model synchronization. The resulting architecture is known as multi-head split learning (MHSL). At the same time, it is important to investigate information leakage, which indicates how much information is gained by the server related to the raw data directly out of the smashed data-the output of the client-side model portion-passed to it by the client. Our empirical studies examine the Resnet-18 and Conv1-D architecture model on the ECG and HAM-10000 datasets under IID data distribution. The results find that SFL provides 1.81% and 2.36% better accuracy than MHSL on the ECG and HAM-10000 datasets, respectively (for cut-layer value set to 1). Analysis of experimentation with various client-side model portions demonstrates that it has an impact on the overall performance. With an increase in layers in the client-side model portion, SFL performance improves while MHSL performance degrades. Experiment results also demonstrate that information leakage provided by mutual information score values in SFL is more than MHSL for ECG and HAM-10000 datasets by 2×10-5 and 4×10-3, respectively.

BiGAMi: Bi-Objective Genetic Algorithm Fitness Function for Feature Selection on Microbiome Datasets.

The relationship between the host and the microbiome, or the assemblage of microorganisms (including bacteria, archaea, fungi, and viruses), has been proven crucial for its health and disease development. The high dimensionality of microbiome datasets has often been addressed as a major difficulty for data analysis, such as the use of machine-learning (ML) and deep-learning (DL) models. Here, we present BiGAMi, a bi-objective genetic algorithm fitness function for feature selection in microbial datasets to train high-performing phenotype classifiers. The proposed fitness function allowed us to build classifiers that outperformed the baseline performance estimated by the original studies by using as few as 0.04% to 2.32% features of the original dataset. In 35 out of 42 performance comparisons between BiGAMi and other feature selection methods evaluated here (sequential forward selection, SelectKBest, and GARS), BiGAMi achieved its results by selecting 6-93% fewer features. This study showed that the application of a bi-objective GA fitness function against microbiome datasets succeeded in selecting small subsets of bacteria whose contribution to understood diseases and the host state was already experimentally proven. Applying this feature selection approach to novel diseases is expected to quickly reveal the microbes most relevant to a specific condition.

Stool and Ruminal Microbiome Components Associated With Methane Emission and Feed Efficiency in Nelore Beef Cattle.

Background: The impact of extreme changes in weather patterns on the economy and human welfare is one of the biggest challenges our civilization faces. From anthropogenic contributions to climate change, reducing the impact of farming activities is a priority since it is responsible for up to 18% of global greenhouse gas emissions. To this end, we tested whether ruminal and stool microbiome components could be used as biomarkers for methane emission and feed efficiency in bovine by studying 52 Brazilian Nelore bulls belonging to two feed intervention treatment groups, that is, conventional and by-product-based diets. Results: We identified a total of 5,693 amplicon sequence variants (ASVs) in the Nelore bulls' microbiomes. A Differential abundance analysis with the ANCOM approach identified 30 bacterial and 15 archaeal ASVs as differentially abundant (DA) among treatment groups. An association analysis using Maaslin2 software and a linear mixed model indicated that bacterial ASVs are linked to the host's residual methane emission (RCH4) and residual feed intake (RFI) phenotype variation, suggesting their potential as targets for interventions or biomarkers. Conclusion: The feed composition induced significant differences in both abundance and richness of ruminal and stool microbial populations in ruminants of the Nelore breed. The industrial by-product-based dietary treatment applied to our experimental groups influenced the microbiome diversity of bacteria and archaea but not of protozoa. ASVs were associated with RCH4 emission and RFI in ruminal and stool microbiomes. While ruminal ASVs were expected to influence CH4 emission and RFI, the relationship of stool taxa, such as Alistipes and Rikenellaceae (gut group RC9), with these traits was not reported before and might be associated with host health due to their link to anti-inflammatory compounds. Overall, the ASVs associated here have the potential to be used as biomarkers for these complex phenotypes.

Putative mobilized colistin resistance genes in the human gut microbiome.

BACKGROUND: The high incidence of bacterial genes that confer resistance to last-resort antibiotics, such as colistin, caused by mobilized colistin resistance (mcr) genes, poses an unprecedented threat to human health. Understanding the spread, evolution, and distribution of such genes among human populations will help in the development of strategies to diminish their occurrence. To tackle this problem, we investigated the distribution and prevalence of potential mcr genes in the human gut microbiome using a set of bioinformatics tools to screen the Unified Human Gastrointestinal Genome (UHGG) collection for the presence, synteny and phylogeny of putative mcr genes, and co-located antibiotic resistance genes. RESULTS: A total of 2079 antibiotic resistance genes (ARGs) were classified as mcr genes in 2046 metagenome assembled genomes (MAGs), distributed across 1596 individuals from 41 countries, of which 215 were identified in plasmidial contigs. The genera that presented the largest number of mcr-like genes were Suterella and Parasuterella. Other potential pathogens carrying mcr genes belonged to the genus Vibrio, Escherichia and Campylobacter. Finally, we identified a total of 22,746 ARGs belonging to 21 different classes in the same 2046 MAGs, suggesting multi-resistance potential in the corresponding bacterial strains, increasing the concern of ARGs impact in the clinical settings. CONCLUSION: This study uncovers the diversity of mcr-like genes in the human gut microbiome. We demonstrated the cosmopolitan distribution of these genes in individuals worldwide and the co-presence of other antibiotic resistance genes, including Extended-spectrum Beta-Lactamases (ESBL). Also, we described mcr-like genes fused to a PAP2-like domain in S. wadsworthensis. These novel sequences increase our knowledge about the diversity and evolution of mcr-like genes. Future research should focus on activity, genetic mobility and a potential colistin resistance in the corresponding strains to experimentally validate those findings.

Nasopharyngeal Microbial Communities of Patients Infected With SARS-CoV-2 That Developed COVID-19.

BACKGROUND: SARS-CoV-2 is an RNA virus causing COVID-19. The clinical characteristics and epidemiology of COVID-19 have been extensively investigated, however, only one study so far focused on the patient's nasopharynx microbiota. In this study we investigated the nasopharynx microbial community of patients that developed different severity levels of COVID-19. We performed 16S ribosomal DNA sequencing from nasopharyngeal swab samples obtained from SARS-CoV-2 positive (56) and negative (18) patients in the province of Alicante (Spain) in their first visit to the hospital. Positive SARS-CoV-2 patients were observed and later categorized in mild (symptomatic without hospitalization), moderate (hospitalization), and severe (admission to ICU). We compared the microbiota diversity and OTU composition among severity groups and built bacterial co-abundance networks for each group. RESULTS: Statistical analysis indicated differences in the nasopharyngeal microbiome of COVID19 patients. 62 OTUs were found exclusively in SARS-CoV-2 positive patients, mostly classified as members of the phylum Bacteroidota (18) and Firmicutes (25). OTUs classified as Prevotella were found to be significantly more abundant in patients that developed more severe COVID-19. Furthermore, co-abundance analysis indicated a loss of network complexity among samples from patients that later developed more severe symptoms. CONCLUSION: Our study shows that the nasopharyngeal microbiome of COVID-19 patients showed differences in the composition of specific OTUs and complexity of co-abundance networks. Taxa with differential abundances among groups could serve as biomarkers for COVID-19 severity. Nevertheless, further studies with larger sample sizes should be conducted to validate these results.