Multi-Peptide Signature as a Pathway Biomarker for Clinical Trial Patient Selection

Was this article helpful?

Patient selection for a pharmaceutical clinical trial is a critical process that considers many biomarkers including molecular, clinical and demographic. Since drugs act on protein pathways, molecular markers are most indicative, but discovering enough protein and metabolite markers is extremely difficult. While mass spectrometry (MS) is known to struggle with intact proteins, it excels at detecting millions of digested peptides from practically every pathway. This suggests MS peptide ensembles can make easier and more sensitive pathway biomarkers than any one biomolecule. 

We developed the AIMS™ technology to compute multi-peptide signatures — from dozens to thousands of pathway peptides — using proteomics MS data. They can be tuned to stratify patients for dose selection, efficacy, liver or kidney risk factors, and other metrics from training data. 

AIMS is a suite of AI-enabled analysis tools to address sensitivity, throughput, cost, and irreproducibility in clinical MS. In contrast to academic proteomics as a spectra-centric statistical science, AIMS views clinical proteomics as a data-driven science centered around intact peptide ions. By using AI trained on public data to reduce the need for fragmentation, MS throughput and cost can be significantly improved for clinical deployment. Some of foundational work was published in well-received conference posters during 2019. 

Why MS protein quantitation is fundamentally irreproducible

Academic proteomics, originally developed to rapidly catalog human proteins, relies on fragment spectra to identify peptides, which are re-assembled by sequence into ‘proteins’. Clinical proteomics adopted academic methods with often irreproducible results.

We identified two causes:

• • Peptide quantity is strongly dependent on sample prep (especially digestion variability), while ‘protein’ quantitation is inherently ambiguous and arbitrary. 
  • Trypsin variability causes non-linear distortion to peptide distribution that cannot be globally normalized. The full solution requires mechanized sample prep (e.g. Pressure Biosciences), rejection of out-of-spec or corrupted data, and peptide-specific normalization. 

Because protein conformation is lost, an inferred ‘protein’ is really a collection of many conformations with indiscernible MS quantitation. If one peptide doubled while another halved with a drug treatment, it is impossible to determine whether the parent protein doubled, halved, or remained constant. Yet most academic software will arbitrarily calculate protein quantities, so that spurious correlations become likely.

Note that academic and clinical research have opposite objectives in minimizing false negatives vs. false positives, respectively. And software has a “no free lunch” principle called Iron Triangle (“Good, fast and cheap — pick any two”). Together they explain why academia-targeted software designed to be fast, inexpensive, and that maximize the quantity of results (i.e. minimize false negatives) tend to produce untrustworthy analyses with false positives. 

Sage-N is an AI app provider 

Give us reference patient MS data with annotated markers (e.g. dosage, efficacy) and we deliver a self-contained app that computes them for new patient MS data, with guaranteed specs. It’s now that simple to screen patients with molecular pathway biomarkers using MS with an app for clinical trials.

For more information, to receive copies of scientific posters, or to schedule a presentation, please contact Terri at Sales@SageNResearch.com.