Using Machine Learning to Count Microscopic Worms

AJAS · 2022 Computer Science

Alyson Michael, Anshul Rastogi, Garima Rastogi

Thumbnail supplied by the source for Using Machine Learning to Count Microscopic Worms

Overview

Caenorhabditis elegans is a prime model organism for biological and toxicological studies. However, counting C. elegans in their various life stages (larva through adult) for research is a time-consuming task. A program was intended to be created that can automate the process of tallying C. elegans by life stage with machine learning. This program is composed of two neural networks: Model 1 and Model 2. Model 1 is an object detector designed to recognize hatched C. elegans and C. elegans eggs in an image. This model was designed with the intention to push the identified images to the next model. Model 2 then classified the identified images into one of three life stages: L1 and L2, L3 and L4, and adult. Images of worm plates taken with an iPhone 6S camera over the ocular lens of a stereo microscope were used to train Model 1. Images were annotated for object detection with Roboflow and their custom YOLOv5 object detection model was trained for 400 epochs with a batch size of 16 in Google Colab. At the end of the training, precision on the validation dataset reached approximately 0.9, while recall reached approximately 0.75. Model 2 was designed as a sequential image classifier made with Tensorflow Keras, and has already been trained successfully on a mock dataset and is yet to be trained on the actual dataset.

Video

If the video does not play, watch it on the original project page.

Garima’s story

I learned about AJAS through NHAS in 2019, where I first participated in the three week research program they offered. That year, I worked with a lab partner and conducted genetic analysis on showy lady's slippers. I learned how to extract DNA and work with related machinery, such as gel electrophoresis machines. This was an amazing experience; for the first time, I watched science I had only ever read about come alive, and I immersed myself in it. Once the program ended, we wrote a paper summarizing the work my lab partner and I did. We were invited to present our work at the AJAS conference in 2020.

I absolutely loved the conference and the surrounding events. That experience inspired me to participate in NHAS's summer research program again, and this year I knew I wanted to work on a computer science project, just to get a taste of another field of science. So, we began this project of cataloging worms using machine learning.

Under our mentor's guidance and with the help of courses, I began from the basics of machine learning and built all the way upto making and training models. We spent three weeks in the lab working on both our models. We also collected and annotated images of C. elegans for our first model, and then trained Model 1. Once that ran successfully, we worked on Model 2 and after much debugging, finally created the framework for the model.

Anshul’s Story:

My journey to AJAS this year begins with the 3-week research program offered by our state’s academy of science (NHAS). I had the opportunity to present at the Spring 2020 AJAS conference through this program in 2019 and had been inducted as an AJAS lifetime fellow at that time. The program was one of my first times working hands-on in a laboratory environment, and the experience was exciting. I chose a project relating to C. elegans that year and learned a lot, including persistence and presentation skills, among many other things.

I participated in the NHAS program again in 2021 and this time, I chose to combine my previous experience with C. elegans with my love of computer science and undertook – with my sister – a project suggested by our mentor: using machine learning to identify and classify C. elegans in images.

This was an opportunity to work on a fascinating research problem while understanding machine learning, which excited me in particular. Machine learning has great potential, and I am excited to see how it continues to shape humanity’s technological progress; it is a field that could have an incredible impact on the world. As a simple example, the NHTSA found during a study in 2018 that 94-96% of collisions are caused by human error. Machine learning is the key to automating driving, which could save hundreds of thousands lives if completed successfully.

After our 3 weeks of research in the NHAS program ended, we wrote a summary paper for the project and were invited to present at AJAS once again. We eagerly look forward to this experience; AJAS has been an inspiring and humbling experience that provides us with an opportunity to discover the work of their peers and scientists in the fields that we aspire to.