Join us virtually on Wednesday, June 9, at 4:30 PM to learn more about quite interesting bioinformatics subjects. You will have the pleasure to hear,
- Éric Audemard, How to improve biomarkers selection using predictive genes
- Nadia Tahiri, Quantitative Structure-Activity Relationship (QSAR) Modeling to Predict the Transfer of Environmental Chemicals across the Placenta
Éric Audemard completed his PhD at INRA Toulouse, creating a new method to detect tandem duplication using graph theory. His postdoctoral research at McGill consisted in the analysis and development of tools associated with cancer and genetic diseases. He currently works at IRIC. More information about EPCY, the novel tool he will be presenting is available on the GitHub Project’s Page.
Dr. Nadia Tahiri obtained her doctorate in Computer Science from the University of Quebec in Montreal in 2018. She is finalizing her postdoctoral research in machine learning at the University of Montreal. Her research interests are in the field of Artificial Intelligence in Public Health, Phylogeny, Phylogeography, and Classification. She is very involved in community initiatives that promote women in technology and is one of MonBUG’s co-organizer.
Abstracts for the two presentations can be found below.
We will use jitsi for the meeting (see the link on the Meetup Page). The event will also be streamed on YouTube Live. You can register for the meeting on our Meetup Page, or by sending an email to info@monbug.ca.
Be sure to contact us via Meetup or email if you would like to present to a future event. We are always looking for enthusiast and enthusiastic presenters.
Éric Audemard, How to improve biomarkers selection using predictive genes
The accurate detection of predictive genes biomarker from high throughput technologies such as RNA sequencing has become an important, yet challenging task. Currently, most of the available methods are based on statistical tests to select Differentially Expressed Genes (DEG). However, to maximize the probability of success and to limit time and resource-consuming validations, candidate genes need to be selected based on criteria in line with the final objective, which is their predictive capabilities. To rank best Predictive Genes (PG) we develop EPCY, a method based on servals classifiers trained by cross-validation. Using both bulk (Leucegene dataset) and single-cell (10X dataset) RNA sequencing data, we demonstrated how this method allows the selection of the best candidates compared to benchmark DEG analysis-based methods. More specifically, we demonstrated the stability of EPCY analysis when the number of cases composing the dataset varies.
Nadia Tahiri, Quantitative Structure-Activity Relationship (QSAR) Modeling to Predict the Transfer of Environmental Chemicals across the Placenta
The increasing diversity of environmental chemicals in the environment, some of which may be developmental toxicants, is a public health concern. The aim of this work was to contribute to the development of rapid and effective methods to assess prenatal exposure. Quantitative structure-activity relationships (QSAR) modeling has emerged as a promising method in the development of a predictive model for the placental transfer of contaminants. Fetal to maternal plasma or serum concentration ratios for 105 chemicals were extracted from the literature, and 214 molecular descriptors were generated for each of these chemicals. Ten predictive models were built using Molecular Operating Environment (MOE) software, and the Python and R programming languages. Training and test datasets were used, respectively, to build and validate the models. The Applicability Domain Tool v1.0 was used to determine the applicability domain. The models developed with the partial least squares regression method in MOE and SuperLearner in R, showed the best precision and predictivity, with internal coefficients of determination (R2) of 0.88 and 0.82, cross-validated R2s of 0.72 and 0.57, and external R2s of 0.73 and 0.74, respectively. The inclusion of all test chemicals by the domain of applicability demonstrated the reliability and relevance of the model predictions. The results obtained demonstrate that QSAR modeling can help quantify the placental transfer of environmental chemicals.