BioTechniques News
Maddy Chapman
Virtual cells based on 4D lattice light-sheet microscopy predict mitochondrial responses to drugs, potentially accelerating therapeutics discovery.
Researchers at the University of California San Diego (CA, USA) have developed two digital models that can predict cellular health and drug responses based on mitochondrial morphology changes. The models could mean we rely less on time-consuming lab experiments in the early stages of drug discovery, accelerating the process for a variety of diseases.
Mitochondria are constantly moving around the cell via dynamic networks, ensuring energy is delivered where its most needed. However, these mitochondrial networks deform in diseased states, which means they can be used as disease markers. Figuring out exactly how these network changes affect cellular health has been limited due to static, 2D snapshots of the system. Now, the development of two digital models is looking to remedy that.
San Diego researchers have published two papers, each detailing an approach to quantify mitochondrial changes in response to drugs. Both digital approaches were developed using single-cell lattice light-sheet microscopy movies, which can capture the three-dimensional movement of mitochondria over time.
The first approach – an AI model called MitoSpace [1] – was trained on 40,000 of these 4D videos of cancer cells treated with 25 mitochondria-perturbing compounds. The model grouped cells that responded similarly to compounds, based on mitochondrial shape alone, with almost 75% accuracy. This is a significant improvement on the 56% accuracy of AI models trained on 2D images, which are commonly used in drug screens.
AI-enhanced antibiotic discovery could thwart multi-drug resistant gonorrhea
AI-enabled antibiotic discovery proves effective at identifying new chemical structures and targets in the constant fight against antibiotic-resistant gonorrhea.
The second approach used a 4D video to create a ‘digital twin’ of a human head and neck squamous carcinoma cell, devising a set of rules to which the virtual cell must adhere [2]. They did this by mapping the mitochondrial networks down to their motor proteins and microtubule tracks on a specialized image-analysis software. The team adjusted and readjusted these networks to ensure they mimicked real mitochondrial transport within human cells.
“We have built a physics‑based virtual cell and can compare it side‑by‑side to the actual 3D microscopy movie, something that has never been possible before,” commented senior author on both papers Johannes Schöneberg. When the team asked the model to predict how it would respond when treated with nocodazole – a microtubule disruptor – it closely mimicked how real cells would respond, demonstrating reduced rate of mitochondrial motion, fusion and fission.
Both models have shown significant potential for predicting cell health and drug response, without having to test cells experimentally in the lab. Looking to the future, the team plans to combine the two digital approaches to create a workflow that can first scour large amounts of data to identify cellular patterns before interrogating the physical reasons for those patterns. Expanding the models to include other organelles will mean they can study whole virtual cells and their responses, which would make them excellent tools for other applications within cell biology.
“The ultimate future is not one cell but multiple cells acting as tissues,” Schöneberg added. “Modeling whole tissues will let us simulate more realistic human biology and eventually inform clinical treatment.”
The post Could AI and ‘digital twin’ models of mitochondria accelerate drug discovery? appeared first on BioTechniques.
Powered by WPeMatico