Category: news

How accurately can we reconstruct microbial interactions from microbiome sequencing data? In our recently published study in PLOS Computational Biology, we address this question through a comprehensive benchmark of 18 computational methods for inferring microbial co-occurrence networks.

Since reliable experimental ground truth for microbial interactions is difficult to obtain, we used synthetic 16S rDNA-seq datasets generated with known underlying interaction networks. Using the N2SIMBA simulator, we designed 189 different simulation scenarios, resulting in a total of 9,450 datasets. These scenarios covered a broad range of experimental and ecological conditions, including different numbers of samples and microbial species, sequencing depths, community diversity levels, network structures, and proportions of competitive interactions.

We evaluated the methods from several perspectives, assessing their ability to identify individual interactions, recover the centrality of key species, reconstruct the overall network topology, and their computational requirements.

Our results reveal important limitations in the current co-occurrence-based approach. Although some methods perform well under favourable conditions, their accuracy decreases considerably in more challenging and realistic scenarios. In particular, performance is strongly influenced by the number of samples relative to the number of species, network density, sequencing depth, and the prevalence of competitive interactions. Overall, conditional-dependence methods showed the most consistent performance, with SPRING achieving the best overall results, followed by CCLasso and COZINE.

Our findings highlight the importance of carefully evaluating and interpreting microbial networks inferred from sequencing data. While co-occurrence analysis remains a valuable tool for generating ecological hypotheses, recovering the true interaction structure of a microbial community is a much more challenging task than simple statistical association. Further methodological and computational developments are needed to improve the reliability of microbial interaction inference.

Congratulations to Piero Mariotto, Matteo Baldan, Barbara Di Camillo and Giacomo Baruzzo for this work!

The article, together with the benchmark datasets, results, and software resources, is openly available:

Read the article: PLOS Computational Biology

Explore the data and computational resources: Zenodo repository

Access the source code: GitLab repository

This work was made possible by the support of the European Union – NextGenerationEU, and of our partner company, EuBiome S.r.l.

🧬 Our group had a great time participating in BITS2026, the 22nd Annual Conference of the Bioinformatics Italian Society, held in Padua from May 27–29, 2026.

The conference provided an excellent opportunity to share our latest research, exchange ideas with colleagues, and connect with the vibrant Italian and international bioinformatics community.

A special highlight was the 2nd International BioHackathon, organized by our group members Giulia Cesaro and Giacomo Bruzzo in collaboration with Young BITS, CINI Young InfoLife, and ISCB RSG Italy. Bringing together talented young scientists from across Italy and Europe, the BioHackathon challenged participants to decode biological networks from omics data through collaboration, creativity, and innovation. The event showcased the enthusiasm and scientific excellence of the next generation of computational biologists.

🎤 We were also proud to see our PhD student Matteo Andriolo present our work on ArchetypalC, a scalable and high-performance C++ implementation of Archetypal Analysis for large-scale bioinformatics applications. ArchetypalC supports sparse matrices, parallel computation, and interactive visualization, enabling efficient exploration of complex biological datasets.

Our group also contributed with two additional posters:

🎤 Elena Marinello presented GRAPHIA, a pipeline for deriving patient-specific network features by weighting a shared biological interaction graph according to individual gene expression profiles.

🎤 Mikele Milia presented scTransformer, a novel framework for SingleCellTranscriptomics that incorporates prior biological knowledge into the self-attention mechanism of Transformer models.

We’re excited to share that our latest paper is now published in Frontiers in Bioinformatics: “N2SIMBA: from Network topology to SIMulation of interactions and BActerial abundance, using microbial consumer resource model.”

In this work, we introduce N2SIMBA, a modular simulation framework designed to generate realistic 16S rDNA-seq count tables starting from known microbial interaction networks. Since experimentally validated ground-truth interaction networks are rarely available, especially for complex bacterial communities, N2SIMBA provides a flexible in silico tool to systematically benchmark network inference methods.

The framework leverages the Microbial Consumer Resource Model to simulate microbial community dynamics through metabolite-mediated interactions, accounting for both competitive and promotional effects between taxa. By combining ecological modelling with sequencing count simulation, N2SIMBA enables users to study how community composition changes under different network topologies, environmental conditions, and perturbations.

We hope this framework will be useful for researchers interested in microbial ecology, systems biology, and the design of robust bacterial communities. New features and extensions are already under active development!.

Congratulations to Matteo Baldan, Giacomo Baruzzo, Piero Mariotto, Ada Rossato, Marco Cappellato, and Barbara Di Camillo for this work.

The paper is available at: https://doi.org/10.3389/fbinf.2026.1783447