How Selvita uses KNIME Data Apps to Make ML Models Accessible to Scientists
Read how Selvita uses KNIME data apps to help chemists and biologists easily access Python-based models.
Cut down manual experimentationsaving time and resources on compound synthesis
Selvita is a publicly listed research organization that provides drug discovery and drug development services for biotech companies, pharmaceutical companies, and academic partners. With a focus on integrated drug discovery projects, efficient collaboration between chemists and biologists is at the center of every project.
Summary
The challenge:
- Scope for improving the speed and effectiveness of drug discovery.
The solution:
- Implemented KNIME as a user-friendly front-end for Python-based Quantitative Structure Activity Relationship (QSAR) and Absorption-Distribution-Metabolism-Excretion-Toxicity (ADMET) predictive modeling.
The results:
- Reduced manual experimentation, cutting down time and resources needed for compound synthesis.
- Enhanced accuracy and consistency in predictive modeling.
Challenge: Making Python-based ML models for QSAR and ADMET available to chemists and biologists
Drug discovery demands both precision and speed. Although traditional methods for physically synthesizing compounds can be improved to save time, money, and resources, the bigger challenge is building effective collaboration between chemistry and biology teams.
To support the chemists in their everyday work, Selvita provides machine learning models in Python for QSAR and the prediction of ADMET properties. These models quickly estimate key molecular properties, like solubility or how a compound might behave in the human body without physically synthesizing and testing each molecule.
But the chemists and biologists who depend on these predictions usually aren't Python experts. They needed an intuitive, user-friendly way to access these predictive results without having to deal directly with complicated code.
“Our chemists, biologists — they don’t want to see any kind of a Python prompt…They want something intuitive.” Jörg Wichard, Principal Scientist, Selvita
Solution: KNIME data apps to make Python based ML models easily accessible to chemists and biologists
The AI and ML team used KNIME data apps to act as a front-end solution, giving chemists and biologists in the organization intuitive access to complex Python-based QSAR and ADMET predictive models.
The solution incorporated multiple technologies:
- Data cleansing from public and in-house sources
- Python-based model development
- MLflow for model versioning
- KNIME to deploy the models as data apps that act as a user-friendly front-end via an API
With easy access to these models via KNIME data apps, chemists could input chemical structures and quickly get predictions regarding key properties such as solubility and drastically reduce unnecessary synthesis efforts.
“The more we can predict, the fewer experiments we have to do, the faster we are, the more effective we are, and the more time and money we can save.” Jörg Wichard, Principal Scientist, Selvita
For example, the team built a data app that is deployed as an API for predicting the aqueous solubility of small organic molecules. This allowed them to quickly check if a molecule would dissolve in water without having to actually synthesize it first — saving valuable time and resources.
Results: More accessible ML models for chemists and biologists
Using KNIME, Selvita was able to:
- Focus on essential chemical syntheses through accurate predictive modeling.
- Provide chemists with clear, rapid, and reliable predictions in an intuitive interface, significantly enhancing productivity.
Using KNIME saves Selvita time, money and material, increasing throughput and decreasing logistical effort.
Learn more about KNIME Business Hub or schedule a call with our life sciences team.