Machine learning

Artificial intelligence is causing a paradigm shift in almost all fields of science and business, an especially in biology and human health. In microbiome science, artificial intelligence is often developed through machine learning techniques. As microbiome datasets increase in size and are increasingly accompanied by richer metadata and multi-omics measurements, they become more suitable for machine learning.  

At Clinical Microbiomics, we help clients utilize the power of artificial intelligence to maximise biological insights gained from their datasets through tools such as: 

  • Response prediction: Identifying which participants are expected to respond to treatment based on their baseline microbiome. Understandinging why some people might not respond to an intervention is one step towards personalized medicine and can be used to guide drug development and better clinical trials. 
  • Biomarker identification: Machine learning is a powerful tool for biomarker discovery due to its ability to detect non-linear relationships between multiple variables and clinical measurements. Mining and prioritising biomarkers from big datasets, followed by validation in independent cohorts before going to the clinic dramatically improves the biomarkers’ chance of success and reduces downstream costs. 
  • Health metrics for quantitative evaluation of treatment: Since it is not known what constitutes a healthy microbiome, effects of an intervention on the microbiome and host health can be difficult to interpret. We have created a set of health metrics based on machine learning, such as our “age predictor” suited to evaluate the effect of treatments or other factors on the microbiome in a quantitative way. 

For microbiome studies with 200 or more samples, researchers may find interesting insights from complementing our standard statistical analysis with machine learning. While machine learning is becoming faster and more accessible overall, analysis of microbiome data requires extra attention due to the nature of the data (high dimensionality, compositionality, sparsity, etc). We have optimised our methods for microbiome and metabolome analysis with a systems-biology approach. This not only allows us to build the most accurate models, but also enables better interpretation since a detailed understanding is often more appropriate than a ‘black-box’ model in life sciences.

Our specialized “foundation” model is trained in a self-supervised manner, which dramatically increases the available training data. This is a general-purpose microbiome model that can subsequently be fine-tuned for more specialized tasks. Given the model’s understanding of dynamics and interactions between microbial species and metabolites, it is able to predict personal responses to an intervention and changes in the microbiome when it is perturbed. 

Contact us to discuss your microbiome research project and learn how we can support.

Get in touch

Tell us about your study.

Please enable JavaScript in your browser to complete this form.