Multi-Omics Factor Analysis for Autoimmune Trials: A Simulated Example

This is a simulated analysis, not a client result. It uses a mock Phase II trial with synthetic data to show the analytical approach. All patient IDs, biomarker values and outcomes are representative and do not come from a real trial.

Key Takeaways

• MOFA+ distils thousands of multi-omics features into a handful of interpretable factors - each representing a meaningful source of biological variation across patients
• On this synthetic dataset, using MOFA factors as inputs to XGBoost moved the high-responder prediction F1-score from 0.667 (full omics) to 0.698, illustrating how dimensionality reduction can support downstream machine learning
• On this dataset the pipeline, from data preprocessing to model interpretation, runs in under a minute on Sonrai Discovery

Jump to section

Simulation Overview

Mock Trial Overview: A Phase II trial evaluating a biologic therapy for rheumatoid arthritis, with Disease Activity Score for 28 joints based on the C-reactive protein level (DAS28-CRP) change from baseline at 6 months as the primary endpoint, supported by flow cytometry and proteomics data collected at baseline and 1-month follow-up.

Challenges of Clinical Data Standardization

Multi-omics clinical trial data presents three recurring challenges that conventional analytical approaches struggle to address:

• Overwhelming Data Complexity. Managing and analyzing thousands of features across multiple omics datasets is time-consuming and error-prone without a unified analytical framework.
• Elusive Biomarkers. Pinpointing robust and meaningful biomarkers for patient stratification can feel like finding a needle in a haystack when individual features are assessed in isolation.
• Fragmented Data Insights. The lack of a unified environment for seamless integration and exploration of multi-omics data hampers the ability to uncover correlations between biological signals and clinical variables.

Traditional clinical trials often focus on single omics layers to identify disease-related genes or predict drug efficacy. This approach frequently fails to capture the complexity of biological systems, where multiple molecular processes interact to influence disease progression and treatment response. Multi-omics addresses this limitation by integrating data from different levels, providing a more holistic view of the underlying biology - and enabling identification of key molecular drivers, novel biomarker candidates, and more targeted therapies for autoimmune diseases such as rheumatoid arthritis. This approach complements single-modality workflows like our T cell response immune profiling and AI-powered single-cell analysis use cases.

The Scenario

In this simulated scenario, a clinical-stage company has a pipeline biologic undergoing Phase II clinical trial. The therapy appeared to work better in certain patients. Identifying these patient subgroups could improve trial outcomes and shape future R&D efforts, ultimately benefiting more patients. They sought a way to maximise insights by integrating multi-omics data - proteomics and flow cytometry - and analyzing it alongside their clinical observations, within a streamlined and efficient framework.

Their pain points were clear:

• They needed a way to distil thousands of omics features into actionable factors
• They wanted to correlate these factors with clinical outcomes but lacked the computational infrastructure and analytical tools to do so
• Manual exploration of individual data points wasn't feasible within their timelines

Strategy

What is Multi-Omics Factor Analysis (MOFA)?

Multi-Omics Factor Analysis (MOFA+) [1-3] is an unsupervised machine learning method designed to integrate and analyse diverse biological data types - including genomics, proteomics, and transcriptomics - collected from the same set of samples. By identifying shared patterns across these datasets, MOFA reduces complex data into a few key factors that represent the main sources of variation. This simplification aids researchers in uncovering underlying biological processes and identifying disease subgroups, thereby enhancing understanding of complex biological systems. The full MOFA+ implementation is available via GitHub and in our bioinformatics tools guide.

Step 1: Data Preprocessing

Ensure that the multi-omics datasets, including flow cytometry and proteomics data, are structured appropriately for MOFA. This typically involves organising the data into matrices where rows represent features (e.g. cell types, proteins) and columns represent samples (patients) (Figure 1).

To assess dynamic changes over time, compute the fold change for each feature by dividing the follow-up measurement by its corresponding baseline value. This fold change metric offers a standardised view of expression shifts, enabling meaningful cross-patient comparisons. Apply appropriate normalisation techniques to correct for technical noise and disparities inherent to each dataset type - common approaches include log-transformation, scaling, or other statistical adjustments to harmonise the data across modalities.

Step 2: Model Training

Once the datasets are ready, input the preprocessed data into the MOFA framework (Figure 2). Specify the number of latent factors to be inferred - these represent the underlying sources of variation across the datasets.

Execute the MOFA algorithm to learn the latent factors (Figure 3). During this process, MOFA identifies patterns and correlations within and between the different omics datasets, effectively capturing the principal components explaining the observed variability. This process runs in an Amazon SageMaker notebook instance on the Sonrai Discovery platform and takes less than a minute to complete for a typical clinical trial biomarker dataset.

Step 3: Model Interpretation

Once MOFA learns the factors, perform a series of exploratory analyses to understand their biological significance, and examine whether any are associated with clinical outcomes such as DAS28-CRP change from baseline. The multi-omic features contributing to the clinically associated factors are interesting to follow up as:

• Biomarkers for patient stratification
• Predictors of treatment response
• Candidates for mechanism of action studies
• New treatment targets for R&D

Step 4: Machine Learning to Predict High Responders

MOFA simplifies complex multi-omics data by reducing hundreds or thousands of features into a handful of factors that capture the most meaningful information. These factors can then be used as inputs for machine learning algorithms to address specific questions. Here, we predict which patients are likely to be "high responders" - defined as patients who improved DAS28-CRP by ≥1.6 after treatment.

Two approaches are compared: one using the original concatenated multi-omics dataset, and the other using the 15 MOFA-derived factors.

These scores describe this synthetic example and do not establish performance on real clinical trial data.

Using full omics dataset: with all raw features concatenated, we trained a classifier using no-code XGBoost on the Sonrai Discovery platform. After testing several hyperparameter combinations, the model achieved a test F1-score of 0.667.

Using MOFA factors: by leveraging the simplified MOFA factors, the same XGBoost workflow reached a higher test F1-score of 0.698. In this simulated example, MOFA reduced dimensionality and supported a downstream machine learning workflow.

XGBoost classification using the full concatenated omics dataset.

XGBoost classification using 15 MOFA-derived factors - achieving F1 0.698 vs 0.667 with the full omics set.

Why Work with Sonrai?

Sonrai combines domain expertise, cutting-edge technology, and a customer-first approach to deliver impactful, efficient solutions for multi-omics data analysis. Key advantages include:

• Faster insights. Interactive multi-omics analysis supports rapid identification of candidate biomarkers.
• Risk mitigation. Auditability and traceability support reproducible analysis, directly addressing the reproducibility challenges that plague multi-omics studies.
• Data integrity. A secure, production-ready repository supports seamless multi-omics data integration, collaboration, and version control to maintain robust data governance.
• Clinical data integration. Harmonise, manage, and analyse clinical trial data across omics layers, aligned with industry standards and trial objectives.
• Cost efficiency. Centralised data management within a scalable, governed infrastructure keeps trial data in one place and frees team time for further R&D.

References

1. Argelaguet R, Arnol D, Bredikhin D, et al. MOFA+: a statistical framework for comprehensive integration of multi-modal single-cell data. Genome Biol. 2020;21:111.
2. Argelaguet R, Velten B, Arnol D, et al. Multi-Omics Factor Analysis - a framework for unsupervised integration of multi-omics data sets. Mol Syst Biol. 2018;14:e8124.
3. MOFA: Multi-Omics Factor Analysis. GitHub repository.

No cost

Want to apply MOFA+ to your own multi-omics data?

Challenge Sonrai with your own data. We'll run multi-omics factor analysis on your dataset - at no cost.

Talk to our team

Sonrai acts as an extension of your team, combining scientific expertise, transparent and reproducible analysis, and a governed research environment to generate decision-ready evidence that can be revisited, reproduced, and built upon throughout the life of a programme.