Morphologically realistic biomechanical model of the fly

NeuroMechFly, the first accurate “digital twin” of the fly Drosophila melanogaster, offers a very valuable test site for research promoting biomechanics and biorobotics. This could help pave the way for fly-like robots, such as the one illustrated here. Credit: EPFL

Drosophila’s digital counterpart

“We used two types of data to create NeuroMechFly,” says Professor Pawan Ramdja of the School of Life Sciences at the Federal Polytechnic School of Lausanne (EPFL). “First, we took a real fly and performed a CT scan to build a morphologically realistic biomechanical model. The second source of data was real fly limb movements, obtained using posture assessment software that we have developed over the past couple of years that allows us to accurately track animal movements. ”

Ramdy’s group, working with Professor Auke Ispeert of the EPFL Biorobotics Laboratory, publishes an article today (May 11, 2022) in the journal Natural science methods demonstrating the first-ever accurate “digital double” of the fly Drosophila melanogastercalled “NeuroMechFly”.

Time flies

Drosophila is the most commonly used insect in the life sciences and the long-term focus of Ramdi’s own research, which has worked on digital tracking and modeling of this animal for many years. In 2019 his band published DeepFly3DDeep learning-based motion capture software that uses multiple camera views to quantify motion Drosophila in three-dimensional space.

Continuing in-depth training, in 2021 Ramdya released LiftPose3D, a method of reconstructing 3D animal poses from 2D images taken from a single camera. Such breakthroughs have provided turbulent areas of neuroscience and robotics inspired by animals, tools whose usefulness cannot be overestimated.


A digital model of Drosophila melanogaster called NeuroMechFly. Credit: Pavan Ramdja (EPFL)

In many ways NeuroMechFly represents the culmination of all these efforts. Limited by the morphological and kinematic data of these previous studies, the model has independent computational parts that simulate different parts of the insect’s body. This includes a biomechanical exoskeleton with articulated body parts such as head, legs, wings, abdominal segments, proboscis, antennae, robe (organs that help flies measure their own orientation during flight) and neural network “controllers” with motor output .

Why build a digital duplicate Drosophila?

“How do we know when we understand the system?” Says Ramdja. “One way is to recreate it. We could try to create a robot fly, but it’s much faster and easier to build an imitation animal. So one of the main motivations for this work is to start creating a model that combines what we know about the fly’s nervous system and biomechanics to test whether that’s enough to explain its behavior. ”

“When we conduct experiments, we are often motivated by hypotheses,” he adds. “So far, we have relied on intuition and logic to formulate hypotheses and predictions. But as neuroscience becomes more complex, we are relying more on models that can bring together many interconnected components, play them out, and predict what might happen if you make adjustments here or there. ”

Test bench

NeuroMechFly offers a very valuable testing ground for research that promotes biomechanics and biorobatics, but only if it accurately represents a true animal in the digital environment. Checking this was one of the main concerns of the researchers. “We’ve conducted testing experiments that show we can replicate the behavior of a real animal,” Ramdia says.

Researchers have for the first time conducted 3D measurements of real flies that walk and shear. They then replicated this behavior using the NeuroMechFly biomechanical exoskeleton in a physics-based simulation environment.

NeuroMechFly Research Group

Jonathan Arregit, Victor Lobata Rias, Auke Ijspeert, Pavan Ramdia, Shravan Tata Ramalingasetti and Gizem Ezdil. Author: Alain Duke (EPFL)

As shown in the article, the model can actually predict various motion parameters that would otherwise not be measured, such as leg torques and ground contact reaction forces. Finally, they were able to use all the neuromechanical capabilities of NeuroMechFly to detect neural network and muscle parameters that allow flies to “run” in ways optimized for both speed and stability.

“These case studies have strengthened our confidence in the model,” Ramdja says. “But we’re most interested in when modeling can’t replicate animal behavior by showing ways to improve the model.” Thus, NeuroMechFly is a powerful testing ground for building an understanding of how behavior arises from interactions between complex neuromechanical systems and their physical environment.

Joint efforts

Ramdja emphasizes that NeuroMechFly has been and will remain a public project. Therefore, the software is open source and freely available to scholars for use and modification. “We have created a tool not only for ourselves but also for others. So we made it open source and modular, and give recommendations on how to use and modify it. “

“More and more progress in science depends on the efforts of the community,” he adds. It is important for the community to use the model and improve it. But one of the things NeuroMechFly is already doing is raising the bar. Previously, because the models were not very realistic, we did not ask how they could be suggested by the data. Here we have shown how this can be done; you can take this model, reproduce the behavior and infer significant information. So I think it’s a big step forward. “

Help: “NeuroMechFly, an adult neuromechanical model Drosophila melanogaster”Victor Labato Ríos, Shravan Tata Ramalingasetti, Pembe Gizem Ozdil, Jonathan Arregit, Auke Jan Ijspeert and Pavana Ramdia, 11 May 2022, Natural science methods.
DOI: 10.1038 / s41592-022-01466-7

Back to top button