Luc T.

Biomarker measurements have become an essential component of oncology drug development, particularly so in this era of precision medicine. Such measurements ensure that clinical studies are testing biological hypotheses and can help make decisions necessary to choose which drugs to continue or stop developing. For those drugs taken forward, biomarker measurements may also help choose the appropriate dose, schedule and patient population. In this context, SANOFI, a French multinational pharmaceutical and healthcare company headquartered in Paris, has been prioritizing a diverse and fast-growing pipeline in Oncology.

PROJECT DESCRIPTION

Leveraging statistical and machine learning approaches in drug discovery and development include target identification and validation, small-molecule design and optimization, prediction of biomarkers, and computational pathology

- Obtain, merge, reshape, clean, ensure and sustain data quality according to good clinical practice

-Handle missing data with multiple imputation and deep learning

-Use descriptive statistics, predictive analytics, machine learning, and other methods to learn about and derive insights from the patterns and relations within a dataset

-mRNA, proteomics and seqRNA expression analysis for biomarker discovery (with R Bioconductor and BioPython)

-Use inferential statistics to learn about the causal relations between variables in a dataset, including randomized field evaluations

-Perform machine learning (including deep learning) with Pycaret, MLFlow, AutoML(H2O), sklearn,TensorFlow, keras,Torch. . . to explore and model data obtained from clinical trials (survey data, genomics, transcriptomics, proteomics, pathology images, ...)

-Assess transcriptomic and proteomic signatures to predict endometriosis using diferent statistical and machine learning approaches (pipeline for feature engineering, feature selection, generate marker combinations and optimize them to obtain the best signature with given biomarkers or/and according to predefined objectives)

-Create intuitive and compelling graphics to visualize and think about data, as well as for reporting

- Modelling longitudinal data, time series analysis, and forecasting

-Use linear and non-linear mixed models to handle missing data

-Analysis of variances, matrix decomposition, parameters estimation, develop models to estimate variance components for longitudinal data on the framework of random regression and structured independence models (with Asreml, BLUPf90 (GWAS, single step)

-Evaluate the potential of genomic information in predicting the phenotypes

-Apply machine learning approaches for clustering and identifying diferent genetic profiles overtime

-Develop Structure antedependence (SAD) models and implemented them in ASREML