Confidence Beyond Compliance
PharmaLex uses state-of-the-art Artificial Intelligence (AI) and its long experience in data science to solve our specific customer’s problems, as an addition to our strategical statistical services. Our team, composed of experts from various scientific fields such as mathematics, Bayesian statistics, chemistry, bioengineering, etc., is working in close relationship with our customer to meet their needs. At PharmaLex, we believe a multi-disciplinary team is the best way to make effective data science and support effective decision making in the pharmaceutical world.
Our healthcare expertise combined with our multidisciplinary team are used to deeply understand our customer’s very questions and envisage top-notch AI solution wherever needed. For instance, it is common to apply AI and machine learning tools on big and unstructured data, as data lake and other LIMS.
The PharmaLex team masters most advanced programming languages that are used in statistics and Data Science such as SAS, R, STAN, JAGS, Python, Java, C++, but also modern operating environments such as Hadoop or Tensorflow (Keras).
Our key value-added deliverable has always been in delivering fit-for-purpose modeling on which solutions are build. To ensure compliance, our statistical experts are working closely with our IT and Quality departments. It allows for example the creation of AI-powered software, that is completely compliant with regulatory environment (GAMP5, 21CFR part 11). This accumulated sum of competencies makes us one of the few being able to develop solutions in full accordance with all the (Bio-)Pharmaceutical regulation (ICH-USP-EMA). In addition, the software validation ensures that the right diagnostic or answer is provided with high confidence. For validating AI solutions, specific trials are developed to ensure fair comparison with human gold standard or other algorithmic solutions. For instance, our clinical experts can evaluate the impact of modeling/prediction uncertainty directly on the outcome of the clinical study.
As most data scientists, we started applying machine learning methodologies in applied pharmaceutical development (process, assay, etc.) using classical tools such as SVM, Random Forest, Lasso, PLS, etc. We developed predictive methodologies to evaluate model quality as the classical metrics may fail at providing the right conclusion. We stay upfront of the technology by using most recent algorithms such as Boosted Trees or by handling uncertainty in Neural Network (similar to what we do in Bayesian statistics). Computational power for handling big data in short time frame is also available.
Examples of applications
- Survival time predictions using big cytometry database: comparison of the performance of various AI algorithms
- Biomarkers detection on large tox datasets
- Root cause analysis based on variable importance metrices from gradient boosting model
- Detection of epileptic seizures
- Automatic quantification of substances from NMR signals or full DAD-chromatogram (even unresolved !)
- Bioprocess control based on RAMAN or NIR spectra
- Kinetic and Dynamic models based on EEG signals
- Chemical process control using NIR spectra
- Statistical deconvolution from cEIF and UV data
- Fraud detection for counterfeit drugs (but also for bank transaction…)