When he first started as a postdoctoral researcher in Eugene Myers's group at Janelia Farm, computer scientist and engineer Parvez Ahammad had to pick up a lot of biology—fast. Janelia's cafeteria soon became his informal classroom; at lunch, he would pepper his colleagues with questions and go back to read more about what he had learned.
Trained in the applied sciences, Ahammad had always been interested in extracting patterns from complex data. Then biology's rich datasets began to capture his attention, and the great need for better algorithms and statistical models in biology—and neuroscience in particular — made the time right for him to enter these fields. "In many cases, experiments are scaling up," he says. "There are a lot of data coming through, and you need to find certain underlying patterns across these multiple experiments."
Ahammad had wanted to be an engineer for as long as he can remember. But after earning a master's degree in electrical engineering at the University of Central Florida in Orlando— here he specialized in signal processing, an approach that aims to tame noisy signals among other things—he found he wasn't entirely captivated by the problems he was working on.
That changed when, as a first-year Ph.D. student at the University of California, Berkeley, Ahammad took a class in computer vision, a field that deals with how machines "see" and extract information from videos and other visual input. "The idea of looking at the world through the eyes of a machine was very appealing," Ahammad says.
As part of his doctoral work under S. Shankar Sastry, Ahammad helped develop a spatial gene expression atlas of the common fruit fly, Drosophila melanogaster. Because scientists could tag only a few genes within a single fly, they had to find a way to assemble data from many flies, accounting for their individual differences in shapes and size to generate a best estimate. The method Ahammad helped develop automatically learns tissue shapes and automatically aligns, extracts, and scores new images, which are then analyzed for many genes. With the new tools, the research group was able to identify genes with similar expression patterns.
In a separate project, Ahammad worked with a colleague from Singapore to develop an algorithm that recognizes actions in videos. When people look for videos online, they normally search using text, but Ahammad wanted to see whether information extracted from video clips of an action—say, jumping jacks—could be used to find similar videos of the motion. To do this, the researchers computed what is called an "optical flow," which is the motion signature between one frame of the video and the next. Using compressed video, they could estimate the optical flow and compute the actions quickly. "Let's say someone is doing a moonwalk, and you don't know what to call it," Ahammad says. "With a little piece of video, you can go into a large database and find someone doing something similar."
It was a natural decision to work with Myers, a bioinformatician whose computer algorithms are key to genetic sequence analysis. The Myers group has since moved into image analysis, working to build detailed reconstructions of the brains of model organisms such as the fruit fly. As a postdoc, Ahammad developed algorithms for analyzing whole brain connections, and these new tools can be used to pose specific questions about how parts of the brain are involved in information processing. One of his ongoing projects, which he will continue as a Janelia junior fellow, is to develop a detailed, three-dimensional reconstruction of the larval stage of an established model within biology, the flatworm Caenorhabditis elegans, using images collected from electron microscopes.
As a junior fellow, Ahammad is hoping to gain a biologist's perspective in choosing relevant problems. Besides Myers, Ahammad will work closely with two others: Vivek Jayaraman, who studies the neural correlates of fly behavior, and Albert Lee, who studies the brain mechanisms of learning and memory in rats. In particular, Ahammad plans to develop algorithms to extract relevant data from in vivo recordings of neuronal activity and relate that information to the animal's behavior.
Looking back, Ahammad had never imagined he would be working at the intersection of neurobiology and computer science. He feels lucky to have worked at Janelia and in other environments that nurtured his interests and encouraged multidisciplinary collaborations. "It's important to work on stuff that makes you excited. If you wake up every day thinking 'Wow, this research problem is really interesting,' it makes it fun to go to work," he says.