Identifying an optimal perturbation to induce a desired cell state by generative deep learning.

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Controlling cell states is crucial in biological research.
• Identifying specific perturbations for desired cell state changes is challenging.

Purpose of the Study:

• To present PAIRING (perturbation identifier to induce desired cell states using generative deep learning).
• To identify cellular perturbations that lead to specific, desired cell states.

Main Methods:

• PAIRING embeds cell states into a latent space, decomposing them into basal states and perturbation effects.
• Optimal perturbations are identified by comparing decomposed effects with the desired cell state transition vector.
• The method utilizes generative deep learning for perturbation identification.

Main Results:

• PAIRING successfully identifies perturbations to transform cell states across various transcriptome datasets.
• The tool was used to identify perturbations for reverting colorectal cancer cells to a normal-like state.
• Simulations provided mechanistic insights into gene expression changes induced by perturbations.

Conclusions:

• PAIRING offers a novel approach to identify perturbations for targeted cell state control.
• The method has significant potential for advancing therapeutic development in various biological domains.
• PAIRING provides mechanistic insights into gene expression regulation.