DL in Cancer Driving Mutations & Single-Molecule Microscopy || Novellus|Technion
Details
We will have two lectures (approx. 35-40 minutes each) in Hebrew. Light refreshments will be served before the first lecture.
Estimated schedule:
1800 refreshments at our conference hall (1st floor)
1830 First lecture
1920 Second lecture
First lecture:
Lecturer:
Lior Zimmerman
Machine Learning Researcher
Novellus
Title:
Determination of Cancer Driving Mutations and Their Response to Targeted Therapies using Deep Convolutional Neural Networks
Abstract:
Cancer is caused by mutations in the DNA. However, DNA mutations do not necessarily lead to cancer - although some mutations are known to be extremely oncogenic, the vast majority of mutations have unknown contribution to disease formation. Various methods of sequencing the DNA of tumor cells have been developed over the years and are used to provide treatment recommendations for cancer patients which are tailored to their tumor’s genetic profile. However, current personalized treatment recommendations are only based on well studied mutations and do not take into account uncharacterized mutations, present in the vast majority of tumors. This strategy could have deleterious effects for the patient.
Therefore, the determination of the functional significance of poorly characterized mutations and their response to targeted therapies is essential for developing new drugs and predicting patient responses. In this talk, I will present a platform which is based on a novel Deep Convolutional Neural Networks (DCNN), trained on fluorescence microscopy images of live cells which were incubated with mutated and wildtype gene variants. This neural network is able to accurately predict the oncogenicity of over 300 mutations in 7 different genes and their responses to targeted therapies in a dose dependent manner. Our system's accurate prediction of drug responses, coupled with the annotation of uncharacterized mutations can be leveraged for clinically relevant uses. for example, optimizing the clinical development process and improving patient's treatment recommendations.
Bio:
Lior earned his M.Sc. in Bioinformatics at the Hebrew University. He is an expert in computational biology with a specialization in machine-learning. In his career, he was one of the founding scientists at IgC (a venture funded computational antibody engineering company) where he developed novel algorithms for antibody design. Since joining Novellus last year, Lior has been developing a set of deep learning algorithms for oncogenicity prediction from microscopic images of live cells.
Second lecture:
Lecturer:
Elias Nehme
Departments of Electrical Engineering & Biomedical Engineering
Technion
Title:
Deep learning for dense single-molecule localization microscopy
Abstract:
In conventional microscopy, the spatial resolution of an image is bounded by Abbe’s diffraction limit, corresponding to approximately half the optical wavelength. Over the last decade, super resolution
methods have revolutionized biological imaging, enabling the observation of cellular structures at the nanoscale. These include the popular localization microscopy methods, like photoactivated localization microscopy ((F)PALM) and stochastic optical reconstruction microscopy (STORM). However, despite the great advancement, existing localization microscopy methods are still limited in their acquisition and postprocessing speeds, and in their ability to extract 3D properties of the imaged samples. This talk will cover two of our recent works, which illustrate how deep learning can significantly boost the performance of super-resolution microscopy methods, enabling fast and precise super-resolution in both 2D and 3D.
https://arxiv.org/abs/1801.09631
https://arxiv.org/abs/1906.09957