MIT Biotechnology Group holds its first life sciences poster session
Dozens of students presented on topics ranging from drug delivery to sea robots.
Russel Ismael–The Tech
On May 8, the MIT Biotechnology Group held its first Life Sciences Poster Session event, where dozens of undergraduates presented highlights from their UROPs, ranging from drug delivery to sea robots.
Engineering lipid nanoparticle delivery of mRNA to treat colorectal cancer
Mercedes Randhahn ’27 at the Traverso-Langer Lab is developing more efficient nanoparticle formulations for treating gastrointestinal (GI) metabolic diseases. Her group is building on mRNA COVID-19 vaccines, which teach human cells to make proteins that target the specified virus, inducing a therapeutic effect.
“I’m interested in studying more about gastrointestinal treatments and trying to help more individuals across the world,” Randhahn said to The Tech.
Randhahn stated that her long-term goal is to treat GI diseases through oral drug delivery methods instead of needle injections. Her team wants to transition into minimally invasive treatments like oral delivery, which may make patients more open to therapeutic care. She stated that her project will continue “several years down the road.”
“Right now, we’re trying to optimize our formulation in vitro [outside an organism],” Randhahn said. “Eventually, we will extend this to in vivo [inside an organism] research with a long-term goal of pursuing a clinical trial.”
Machine Learning for aquaculture hatchery production
Unyimeabasi Usua ’27 and Santiago Borrego Garcia Cancho ’26 are part of the MIT Sea Grant, and their research involves how to make aquaculture hatchery production more efficient.
“[Aquaculture hatchery is] basically just these tanks that have thousands and thousands of shellfish larva in them,” Cancho stated in an interview with The Tech. “This is in a hatchery in Cape Cod, and [other Sea Grant researchers] assess the health of these larvae over time and try to help them grow.”
This involves measuring how many larvae have died. To do so, Cancho explains that they view a water sample under a microscope and “count the living and dead ones by hand,” which is an “obviously slow process.” As such, Usua and Cancho’s team is using machine learning to automate the process of identifying live and dead larvae.
They process water sample images to enhance their clarity and accentuate each individual larva for easier counting. Cancho and Usua then trained a “classifier” to specifically look for those larvae, which had an 80% identification rate. Of the 80% of larvae identified, the classifier had a 93% accuracy in determining whether it was dead or alive. As such, the time they saved allowed them to focus more on grouping different larvae.
To classify larvae, Usua stated that they are exploring convolutional neural networks (CNN). She said that CNN works better for the larvae images because it can extrapolate from the images and predict where larvae might be in a non-constrained directive. Before CNN, Usua stated that they identify larvae based on their geometry type.
“Our previous identification process was based on the [larvae] circular shape, but when the hatchery gave us new images with unique shapes in them, our current model wasn’t working,” Usua stated. “CNN would be able to look for individual larva and classify them based on pixels.”
Their next steps are to expand on their model and train it to classify different larvae species.
Synergistic effects of siRNAs and small molecule drugs for nanoparticle delivery systems in ovarian cancer
Valeria Mejia ’27 is looking into how to combine small interfering RNAs (siRNA), which are RNAs that prohibit certain genes from being expressed, with small molecule drugs at the Hammond Lab. siRNAs are noncoding RNAs that prohibit certain genes from being expressed. Mejia stated that siRNAs and small molecule drugs alone have many disadvantages as therapies, as siRNAs can lead to high tumor resistance rates by inducing more mutations, and some small molecule drugs can be highly toxic to patients.
By combining siRNAs with small molecule drugs, Mejia stated that the “synergy” will reduce the disadvantages of both methods.
“The Hammond lab is very much nanoparticle based,” Mejia said to The Tech, so they “introduce these synergistic combinations into nanoparticle systems, and test them out in both in vivo and in vitro experiments.”
Mejia used luciferase-based assays, a method that uses a protein that can activate or deactivate certain genes, to see which cells would respond based on their luminescence. To determine the viability of her cell fibroblasts, she used PrestoBlue assays to compare her cells.
Mejia said she plans to test her methods on mice soon after finishing her in vitro tests. She also stated that her methods would also likely “work in other cancers” in addition to ovarian cancer and is hopeful in its applications.
Predicting gene regulatory functions with deep learning
Jayashabari Shankar* ’27 attempts to “decode human evolution” by using human ancestor regions (HARs) and human ancestor quickly evolved regions (HAQERs) in the human genome.
“There are thousands of regions called [HARs] and [HAQERs], and these regions are key to making us human,” Shabari stated to The Tech. “They distinguish us from other animals like chimpanzees and gorillas because we share 98.8% similarity in terms of genomes.”
Shankar stated that analyzing genome sequences in humans can take years, which is what happened with her postdoctoral research mentor. To accelerate this process, Shabari fed HARs and HAQERs of interest into Enformer, a deep learning model with a library of genes, to predict “what the gene expression will be, where it will act, and what it will do.”
“For instance, for this HAR only expressed in astrocytes [a type of cell in the central nervous system], you can find a lot of specific things,” Shankar said. “If there is a lab that focuses on astrocytes, they’ll be able to find your specific HARs in these astrocytes instead of looking at all 3,000 astrocytes.”
By using Enformer, Shankar’s team is able to streamline the wet lab portion. She also said that since they know what the HARs will do, their work will also help future researchers working on similar topics because her work will be added to the Enformer database. She said her next step would be to analyze thousands more HAQERs in the summer.
“We’ve only done 3,000 when there’s actually, I think, a few million HAQERs,” Shankar clarified. “It’s all a matter of putting it into a supercomputer and analyzing that, but it’s going to be a summer project for me, for sure.”
*Shankar is one of The Tech’s News Editors. She was not involved in this article’s publication.