Netramark presents AI findings in depression trials

Executive Summary

• Netramark Holdings Inc. presented two significant applications of its NetraAI technology at the ISCTM Autumn conference and the ECNP Congress, demonstrating advancements in precision psychiatry and clinical trial design.
• The first presentation analyzed ketamine trials for major depressive disorder (MDD), revealing that NetraAI can distinguish true pharmacologic effects from placebo effects by identifying distinct patient responder subgroups that diverge completely from placebo responders by the second infusion.
• The second presentation focused on the CAN-BIND trial for escitalopram, where NetraAI improved predictive accuracy by reducing clinical variables from 718 to 8 key features and identified a genetic signature (hypomethylation at three gene sites) that predicted treatment response with 91% accuracy.

Key Details

• Conference Presentations:
  • Presented at the International Society for CNS Clinical Trials and Methodology (ISCTM) Autumn conference.
  • Presented at the European College of Neuropsychopharmacology (ECNP) Congress.
• Ketamine Trial Analysis (ISCTM Presentation):
  • Study Type: Randomized, crossover trial of ketamine in treatment-resistant depression.
  • Functional Unblinding: Analysis showed ketamine responders had distinct baseline characteristics compared to placebo responders.
  • Divergence Metrics: Separation between ketamine and placebo responder Personas increased from 0.75 (moderate) after the first infusion to 1.0 (complete separation) by the second infusion.
  • Cohesion Metrics: For ketamine responders, within-group cohesion rose from 25% after infusion 1 to approximately 67% after infusion 2.
  • Implication: Findings indicate ketamine's efficacy is not simply due to functional unblinding/expectancy effects and suggest opportunities for persona-guided trial enrichment and stratified randomization.
• Escitalopram Trial Analysis (ECNP Presentation):
  • Study Source: CAN-BIND trial data.
  • Variable Reduction: Traditional ML applied to 718 variables showed modest accuracy; NetraAI refined the feature set to 8 key variables (covering anhedonia, daily functioning, appetite, and negative thinking).
  • Subgroup Identification: Identified 26 non-responders, 23 responders, and 124 unknowns, leading to improved classifier performance.
  • Genetic Signature: Among 169 patients with ~68,600 methylation variables each, NetraAI identified 33 of 117 responders as highly predictive responders (HPRs).
  • Predictive Accuracy: Hypomethylation at three specific gene sites (LINC01580, STK24, ATXN7L3) linked to neuroplasticity predicted HPRs with 91% accuracy when traditional classifiers were retrained on these features.
  • Mechanistic Insight: The identified genes support synaptic remodelling, chromatin plasticity, and hippocampal neurogenesis.
• Company Technology:
  • NetraAI uses a topology-based algorithm to parse patient data into explainable and unexplainable subsets to prevent overfitting and improve insights from small data sets.

Notable Quotes

• "The ability to distinguish true pharmacologic effects from placebo and to predict treatment response with explainable subgroups represents a meaningful advancement," said Josh Spiegel, president at Netramark. "These findings demonstrate how our technology can transform the way psychiatric and CNS trials are designed, interpreted and translated into clinical practice."