DEEMD: Drug Efficacy Estimation Against SARS-CoV-2 Based on Cell Morphology With Deep Multiple Instance Learning.

Area of Science:

• Computational biology
• Virology
• Drug discovery

Background:

• Drug repurposing offers a faster route to identify SARS-CoV-2 treatments, leveraging existing safety and supply chain data.
• SARS-CoV-2 infection causes significant cellular structure changes, detectable via bioimaging.
• Morphological profiling of cells provides a quantitative method to assess treatment efficacy.

Purpose of the Study:

• To develop a computational pipeline, DEEMD, for identifying potential SARS-CoV-2 antiviral treatments.
• To utilize deep neural networks and morphological profiling for drug screening.
• To validate the DEEMD approach using a public dataset of infected and treated cells.

Main Methods:

• Developed DEEMD, a deep learning pipeline employing multiple instance learning for drug repurposing.
• Extracted morphological features from cell images to create cell morphological profiles.
• Employed weak supervision for infected cell localization, avoiding pixel-level annotations.

Main Results:

• DEEMD successfully identified known SARS-CoV-2 inhibitors, including Remdesivir and Aloxistatin.
• The computational pipeline demonstrated efficacy in estimating treatment effects based on morphological profiles.
• Validated the approach on the RxRx19a dataset, showing its potential for antiviral drug discovery.

Conclusions:

• DEEMD is a validated computational tool for identifying antiviral compounds through morphological analysis.
• The approach can be applied to other emerging viruses and datasets for rapid drug discovery.
• DEEMD facilitates drug repurposing for infectious diseases by analyzing virus-induced cellular changes.