We envision the MEDomics platform becoming a powerful tool that strengthens collaboration among medical AI research teams. This collaborative platform for medical data modeling aims to develop predictive models for personalized treatments. MEDomics features analytical and machine learning capabilities and ultimately helps improve the quality of care provided to patients.
Indeed, MEDomics enables researchers—whether they are clinicians or computer scientists—to manage the entire lifecycle of a healthcare project (data, training, and evaluation) within a single interface. By combining a no-code and a code-based approach, the tool aims to rapidly transform complex clinical data into concrete predictive solutions to improve patient care. These individuals collaborate on defining the clinical problem, collecting data, managing data, generating AI hypotheses, refining hypotheses, and visualizing results.
MEDomics addresses real-world needs:
- Tool fragmentation: The platform centralizes data management, training, and evaluation within a single workspace to avoid the need for multiple software applications.
- Technical barriers: It enables non-programmer researchers to perform predictive modeling in medicine using a code-free graphical interface.
- Collaborative synergy: It facilitates synergy between clinical and IT staff by converting visual workflows into editable Python code.
Key definitions
The name
- Data visualization
- MED = Heterogeneous Medical Data
- Omics = Characterization of biological processes
- Lab = Work platform
- Data integration
- Data pre-processing
- Data analysis
Graphical Pipeline (or Workflow)
A graphical pipeline is a visual representation of a sequence of technical operations. Instead of writing lines of code to load an image, filter it, and then analyze it, the user connects functional “nodes” to one another on a canvas.
Prediction Model (Medical AI)
This model is an algorithm trained on medical data (text, images, or other) to identify complex patterns. Once validated, it can help diagnose a disease or estimate a patient’s prognosis.
Code generation
Code generation is an automated process through which the platform translates a visual diagram (the graphical pipeline) into a structured programming language (Python).
- Researchers
- Research professionals
- Healthcare professionals
- Interdisciplinary synergy: enables clinical staff and IT specialists to collaborate more easily.
- Democratization of medical AI: empowers healthcare professionals to independently lead complex research projects using a no-code interface.
- Customization and reproducibility: ensures the quality and scientific rigor of results through an open-source, modular Python library that can be adapted to the specific needs of each project.
- Model performance: facilitates the optimization of predictive algorithms by enabling a rapid transition between visual configurations and source code.
Here are the features offered by the MEDomics platform:
- Visual pipelines for creating machine learning experiments.
- Unified dashboard: a single interface for importing datasets, monitoring model training, and viewing performance metrics in real time.
- Source code generation: export the visual pipeline to Python scripts (.ipynb).
- Evaluation and interpretation of AI models.
- Deployment of AI models for application on new data.
Case Study: Predicting and Managing Emotional Distress
A primary care team wants to identify patients experiencing emotional distress (anxiety, depression, irritability) using health questionnaires (PROMs/ PREMs) in order to offer psychological support before their condition worsens.
Using MEDomics:
- Analysis of Determinants
- Staff use exploratory tools to correlate emotional distress with other factors in the patient’s life, such as fatigue, chronic pain, or financial insecurity (difficulty paying for meals or rent).
- Automatic identification
- Using the Learning Module, the platform trains a model that learns to identify at-risk patient profiles by combining their responses regarding vitality, sleep, and their sense of involvement in their care.
- Clinical Validation
- Through the Evaluation Module, clinicians verify the reliability of predictions and identify which specific factors (e.g., a low sense of being treated as “a whole person”) most significantly contribute to mental distress.
- Targeted Intervention
- The Application Module generates a prediction for each new completed questionnaire, alerting the team if a patient falls into the “emotional distress” category.
Impact
Patients benefit from a holistic and compassionate approach to their health. Early detection of distress allows for personalized care, facilitates discussions about emotional well-being during appointments, and enables patients to be quickly referred to appropriate social or psychological resources, thereby improving their overall quality of life.
The MEDomics platform is developed by the MEDomicsLab research laboratory (medomicslab.com)
MEDomicsLab. (s. d.). MEDomics.
https://medomics.app