1. Data Ingest
Multi-omics data is received from large, public oncology data sources and private partner data sources. Lantern’s proprietary data is included here as well.
2. Data Preprocessing
This step includes data cleaning, transformation, normalization, and integration without compromising the original quality of data – datasets are reviewed by internal KOLs to ascertain relevance and quality. Data is standardized across patient samples, and connections are made that link preclinical models to patient samples and clinical data.
3. Feature Selection
RADR-A.I. performs proprietary gene filtering via biological, statistical and machine learning-based methods to extract relevant and significant biomarker and genomic features. Findings are benchmarked against published literature and data.
An automated artificial intelligence algorithm uses filtered gene sets to build, tune and test the algorithm in an iterative process that selects an optimal model having high testing accuracy. This module will also generate 50-200 drug tumor-specific candidate biomarkers.
5. Hypothesis Generation, Validation, and Feedback
Data mining and Machine Learning pipelines lead to hypothesis generation, which guides additional preclinical research in the lab, which produces more data. This data can be used to validate hypotheses, which are then added back to the system, thus improving the accuracy of our predictions.
6. Patient Stratification and Clinical Trial Design
Our final, tuned model uses available patient data on candidate biomarkers to predict medical drug response and stratifies patients as responders, partial responders, or non-responders. Response prediction informs both companion diagnostic (CDx) development and clinical trial design for advancements in experimental medicine.