Quetelet seminar

Schedule:

14:30 – 15:15 Dr. Charlotte Soneson Single-cell method benchmarking - current state and future perspectives

15:15 – 16:00 Prof. Davide Risso Stochastic generalized matrix factorization for the scalable analysis of single-cell RNA-seq

16:00 - 17:00 Coffe break and Discussion

Abstracts

Single-cell method benchmarking - current state and future perspectives

Researchers in computational biology are often faced with a choice between a large collection of computational methods for performing a given type of data analysis. Method benchmarking aims to rigorously compare the performance of different methods using well-studied datasets, to characterize and determine the strengths of the respective methods and to provide recommendations to users. In this talk I will discuss the current state of the field of method benchmarking, specifically focusing on methods for single-cell analysis. I will also point to current challenges, and exemplify with some efforts that we have taken over the years to make benchmarking more extensible and continuous.

Stochastic generalized matrix factorization for the scalable analysis of single-cell RNA-seq

Prof. Davide Risso will present a scalable adaptive stochastic gradient descent algorithm tailored for

the estimation of high-dimensional generalized matrix factorization models under generic exponential family likelihoods. Applied to single-cell RNA-seq data, the method outperforms existing implementations of both generalized and non-negative matrix factorization, demonstrating faster execution times while maintaining, or even enhancing, matrix reconstruction fidelity and accuracy in biological signal extraction. The issues of model selection and mini-batch strategy choice will also be discussed.