Results
COMPASS has been applied to biological datasets where the analytical challenge extends beyond identifying individual molecules.
The studies below illustrate two related capabilities: analyzing very large measurement spaces to identify signals associated with a biological outcome, and using those signals—individually and collectively—to characterize meaningful differences between sample populations.
Blinded Treatment-Response Study
Finding treatment-response signals within nearly one million measurements per sample
In a blinded study conducted with a pharmaceutical partner, COMPASS was used to analyze paired plasma samples collected from patients before treatment and following successful clinical response.
Magellan received samples from 10 patients, for a total of 20 samples, and was blinded to the disease indication, therapeutic mechanism, and treatment-response status beyond knowing which samples were paired.
The Analysis
LC-MS analysis generated approximately 945,000 measurable datapoints per sample.
Rather than reducing the dataset to a predetermined list of molecules or selecting candidate features before comparison, the broader measurement space was retained for analysis.
COMPASS compared measurements across the paired sample populations to determine which regions of the data were most strongly associated with the biological difference between the groups.
The Result
From approximately 945,000 datapoints per sample, COMPASS identified approximately 180 features showing complete, non-overlapping separation between the two sample populations.
Nearly all of these features increased following successful treatment, suggesting a coordinated biological response rather than isolated analytical differences.
Importantly, the analysis produced these results while Magellan remained blinded to the disease being studied and the mechanism of the therapy.
From Signal to Biology
The identified features included both peptide and non-peptide signals.
The pharmaceutical partner selected peptide-associated regions for additional investigation using targeted MS/MS. Ten peptides were successfully identified and mapped to their parent proteins.
Following unblinding, those proteins were found to participate in biological processes relevant to the disease and treatment response, including immune activity, fibrosis, and epithelial–mesenchymal transition, as well as functions involving protein degradation and chromatin regulation.
What the Study Demonstrates
The study provides an example of discovery beginning with the broader measurement space rather than a predefined molecular target.
COMPASS reduced a dataset containing nearly one million measurements per sample to a focused group of signals associated with treatment response. Molecular identification could then be directed toward those regions of interest, connecting the computational findings back to relevant biology.
Peer-Reviewed Publication
Differential Serum Peptidomics Reveal Multi-Marker Models That Predict Breast Cancer Progression
Published in Cancers in 2024, this study investigated whether low-molecular-weight serum peptides could distinguish between early- and late-stage breast cancer.
The research compared traditional biomarker discovery with a computational approach using COMPASS to analyze the broader MS1 measurement space.
The Research Question
Breast cancer progression involves complex biological changes that may be reflected in circulating serum molecules.
Rather than assuming that a single biomarker would adequately represent those changes, the study investigated whether patterns across multiple serum measurements could differentiate stage I and stage III breast cancer samples.
Conventional Biomarker Analysis
A traditional feature-based workflow initially identified 65 candidate biomarkers with statistically significant differential expression.
A subset of these markers was subsequently confirmed, and combinations of markers were used to construct predictive models.
The strongest multi-marker models retained predictive information when evaluated against a blinded validation set, demonstrating the potential value of combining information from multiple biological signals.
Whole-Dataset Analysis with COMPASS
The same mass spectrometry files were also analyzed computationally using COMPASS.
Rather than manually selecting individual peaks for analysis, COMPASS comparatively analyzed the entire MS1 dataset from 39 serum samples.
Depending on analytical resolution, this represented approximately 685,000 to 2.5 million datapoints evaluated for their relationship to breast cancer stage.
COMPASS ranked regions of the measurement space according to their ability to distinguish early- and late-stage samples and then incorporated information from multiple datapoints into machine-learning classification.
Within this study cohort, the resulting classifier correctly categorized all subjects when sufficient informative datapoints were incorporated. The study also compared results against randomized group assignments to evaluate how much apparent predictive information could arise from natural variation and overfitting.
From Computational Signal to Molecular Investigation
The study further demonstrated the relationship between feature discovery and molecular identification.
One differentially expressed biomarker was characterized by tandem mass spectrometry as a peptide fragment of fibrinogen alpha chain, a protein with established biological relationships to cancer progression and other physiological processes.
The study illustrates how computational analysis can first identify regions of a complex dataset associated with a biological distinction and subsequently focus molecular characterization on those regions.
AlZaabi A, Piccolo S, Graves S, Hansen M. Differential Serum Peptidomics Reveal Multi-Marker Models That Predict Breast Cancer Progression. Cancers. 2024;16:2365.
What These Studies Demonstrate
Although the biological questions and experimental designs were different, the studies demonstrate several characteristics of the COMPASS analytical approach.
Analysis at Scale
COMPASS can evaluate hundreds of thousands to millions of measurements per sample rather than requiring analysis to begin with a small predefined feature set.
Discovery Before Identification
Signals can be evaluated for their relationship to a biological question before their molecular identities are known, allowing identification efforts to focus on measurements shown to contain useful information.
Population-Based Comparison
COMPASS uses relationships across groups of biological samples to identify measurements associated with disease state, treatment response, or other experimental classifications.
Multiple Signals, One Biological Question
Complex biological differences are rarely represented by a single measurement. COMPASS enables information distributed across many signals to be evaluated collectively.
A Path Toward Integrated Analysis
The same general framework—preserving measurements, identifying relationships across populations, and focusing subsequent interpretation on informative signals—provides a foundation for incorporating MS1, MS2, and complementary omic information within increasingly integrated analyses.