mb-PHENIX: diffusion and supervised uniform manifold approximation for denoizing microbiota data

• Published in: Bioinformatics (Oxford, England) Vol. 39; no. 12
• Main Authors: Padron-Manrique, Cristian, Vázquez-Jiménez, Aarón, Esquivel-Hernandez, Diego Armando, Martinez Lopez, Yoscelina Estrella, Neri-Rosario, Daniel, Sánchez-Castañeda, Jean Paul, Giron-Villalobos, David, Resendis-Antonio, Osbaldo
• Format: Journal Article
• Language: English
• Published: England Oxford Publishing Limited (England) 01.12.2023
• Subjects: Abundance; Algorithms; Approximation; Availability; Data structures; Diffusion; Machine Learning; Manifolds; Microbiota; Microorganisms; Reproducibility of Results; Shotguns
• ISSN: 1367-4811; 1367-4803; 1367-4811
• DOI: 10.1093/bioinformatics/btad706

Microbiota data encounters challenges arising from technical noise and the curse of dimensionality, which affect the reliability of scientific findings. Furthermore, abundance matrices exhibit a zero-inflated distribution due to biological and technical influences. Consequently, there is a growing demand for advanced algorithms that can effectively recover missing taxa while also considering the preservation of data structure. We present mb-PHENIX, an open-source algorithm developed in Python that recovers taxa abundances from the noisy and sparse microbiota data. Our method infers the missing information of count matrix (in 16S microbiota and shotgun studies) by applying imputation via diffusion with supervised Uniform Manifold Approximation Projection (sUMAP) space as initialization. Our hybrid machine learning approach allows to denoise microbiota data, revealing differential abundance microbes among study groups where traditional abundance analysis fails. The mb-PHENIX algorithm is available at https://github.com/resendislab/mb-PHENIX. An easy-to-use implementation is available on Google Colab (see GitHub).