DEEMD

DEEMD applies deep neural networks within a multiple instance learning framework to analyze fluorescence microscopy images and estimate antiviral treatment efficacy for SARS-CoV-2 by quantifying infection-induced morphological changes.

Key Features:

• Deep Neural Networks and Multiple Instance Learning: Employs deep neural network models within a multiple instance learning framework to detect subtle morphological alterations associated with SARS-CoV-2 infection.
• Morphological Feature Extraction: Extracts discriminative features from fluorescence microscopy images to generate detailed morphological profiles for infected and non-infected cells.
• Weak Supervision for Cell Localization: Localizes infected cells using weak supervision without requiring pixel-level annotations.
• Statistical Modeling of Treatment Efficacy: Compares morphological profiles of treated and untreated infected cells to non-infected profiles to estimate treatment efficacy statistically.
• Validation with Known Inhibitors: Has identified known SARS-CoV-2 inhibitors, including Remdesivir and Aloxistatin, as validation of its approach.

Scientific Applications:

• Drug Repurposing for SARS-CoV-2: Enables identification of candidate antiviral compounds by analyzing publicly available datasets such as RxRx19a.
• Quantitative Analysis of Infection-Induced Morphology: Provides quantitative measures of cellular responses to SARS-CoV-2 infection and to drug treatments.
• Application to Emerging Viruses: Can be applied to study morphological responses and identify antiviral treatments for other emerging viruses.

Methodology:

Processes fluorescence microscopy images from the RxRx19a dataset (infected and non-infected, with and without drug treatments); extracts morphological features to create profiles; applies deep neural networks within a multiple instance learning framework and weak supervision to localize infected cells; and uses statistical modeling to estimate treatment efficacy by comparing morphological profiles.