AI Model Predicting Drug Indications for Unknown Substances
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting drug indications in humans are limited, as they can only predict efficacies of substances with known effects, failing to account for substances with unknown efficacies used in training data.
Innovation Solution
A method involving an artificial intelligence model trained with three data sets: one set containing biomarker dynamics from non-human animals administered with known substances, another with substance names linked to their indications, and a third with adverse events, allowing prediction of indications for substances without prior known efficacies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional AI models are trained only with known substance indications, then prediction accuracy for known substances is maintained, but the model cannot predict indications for substances with unknown efficacies
Solution Approach 1:
The patent introduces an intermediary substance with unknown indication as a bridge between known substances. By training the AI model with this intermediary substance that has no known indication, the model learns to predict indications for completely unknown substances while maintaining reliability through the gradual transition from known to unknown indications in the training sequence.
Solution Approach 2:
The patent performs preliminary action by pre-training the AI model with substances of known indications before introducing substances with unknown indications. This staged approach allows the model to first establish baseline prediction capabilities with reliable data, then progressively adapt to predict indications for unknown substances.
2Adaptability or versatility
If more comprehensive training data including unknown efficacy substances is used, then prediction capability is improved, but training data quality and reliability decrease
Solution Approach 1:
The patent segments the training data into distinct categories: substances with known indications, intermediary substances with unknown indications, and test substances. This segmentation allows the model to learn different patterns from each category while maintaining overall data quality, as each segment serves a specific training purpose without contaminating the others.
Solution Approach 2:
The patent changes the indication status parameter of training substances from 'known' to 'unknown' in a controlled manner. By systematically varying this parameter across different training subsets, the model learns to handle both known and unknown indications while maintaining measurement precision through controlled parameter transitions rather than random data quality degradation.
3Reliability
If traditional clinical trial processes are followed, then regulatory approval is achieved, but time and cost are excessively high with >80% dropout rate in phases I-III
Solution Approach 1:
The patent performs preliminary action by using the AI model to predict indications and filter candidate substances before initiating expensive Phase I-III clinical trials. This pre-screening process identifies the most promising substances with highest predicted indication match, thereby reducing the number of substances that need to proceed to costly clinical trials and minimizing the 80% dropout loss.
Solution Approach 2:
The patent enables skipping of certain traditional trial phases by using AI-predicted indication information to accelerate the drug development process. Substances with high-confidence AI predictions may bypass some preliminary screening steps or proceed more quickly through trial phases, reducing overall development time while maintaining regulatory approval probability.
Data Source
AI summary
The present disclosure provides a method for predicting an indication for a test substance in humans, a prediction device for predicting an indication for a test substance in humans, a computer program for predicting an indication for a test substance in humans, and a prediction system for predicting an indication for a test substance in humans.