We develop ad-hoc approaches based on machine learning applied to real-world data, as well as Natural Language Processing (NLP) in order to identify the opportunities for indication extension or targeting a certain disease. This can be leveraged to identify the conditions for which a molecule can have a positive impact on the patients’ outcomes in diseases for which the molecules are not yet indicated. Such insights can guide the life cycle and portfolio management and support drugs repurposing.
Use cases examples include:
- Out-licensing strategy definition
- Portfolio management
- Life cycle management
- Definition of the target value proposition
- Drug effectiveness prediction to support market approval
- Preparation of value-based reimbursement contract
Exploration of new target indications
Context: identifying opportunities for drug repurposing
Licensing activities allow pharmaceutical companies extracting value from past investments while bringing new therapeutic solutions to patients. Quinten Health helped a pharmaceutical company to assess the interest in targeting hearing disorders with the existing pipeline. 10% of the projects pre-selected for out-licensing had potential for use in such disorders. It was important for the client to understand the biological pathways involved in hearing disorders, the druggable targets these pathways contained, as well as licensable candidates which matched such targets. In addition, biological evidence was needed to inform the identification of candidate molecule to treat hearing disorders.
Quinten Health’s solution: NLP-based approach
To reach this objective, Quinten Health performed text-mining of biomedical literature to gather current knowledge on pathways, genes, and cell processes involved in hearing disorders. We constituted a corpus of relevant publications related to hearing disorders. We then identifying entities and their relationships: molecular bindings, secondary pathways, up- and down-regulated genes. The result of this study leveraging Natural Language Processing allowed to better document the mode of action in order to support the decision making, as well as to find scientific rationale and the potential interest in hearing disorders.
As a next step, we worked on the identification and prioritization of potential indications from learnt vectoral representations. New indications were identified and prioritized using vectoral representations of biomedical entities. We then searched for the rationale for new potential indications via vectoral similarities and modelled relationships between entities.
Impact: actionable insights
The output of the project was a ranking of potential indications, known or new, driven by synthetized scientific knowledge, as well as a visual rationale for the use of the molecule of interest for each indication.