The study published last month in the Journal of Urban Health shows that the child populations in some neighborhoods are not being tested as completely as they should be, said Khalifa Afane, a student in the M.S. program in data science who wrote the study with his advisor, Juntao Chen, Ph.D., an assistant professor in the computer and information science department.

For the study, they used the city’s publicly available lead testing data, which he said “nobody has analyzed before” at the neighborhood level.

A Toxic Heavy Metal

Lead is a toxic heavy metal that can cause learning disabilities and behavior problems. Children pick it up from lead-based paint or contaminated dust, soil, and water. Lead exposure risk “remains persistent” among vulnerable groups including low-income and non-Hispanic Black children, the study says.

The city promotes blood lead level testing and awareness of lead poisoning in high-risk communities through a variety of educational efforts and partnerships.

But some high-risk neighborhoods still don’t get enough testing, Afane said.  A case in point is Greenpoint in Brooklyn vs. South Beach in Staten Island. The study says that despite similar numbers of children and similar rates of lead testing, Greenpoint has consistently averaged eight times more cases—97 out of 3,760 tests conducted in 2021, compared to just 12 in South Beach that year (out of 3,720 tests).

There should actually be more testing of children in Greenpoint, Afane said, because their risk is clearly higher. While testing efforts have expanded in the city, he said, “it matters much more where these extra tests were actually conducted,” since lead is more prevalent in some neighborhoods than in others, he said.

More than 400 Cases May Have Been Missed

For the study, he analyzed test result data from 2005 to 2021, focusing on children under 6 years old who were found to have blood lead levels of 5 micrograms per deciliter. Afane applied a machine learning algorithm to the testing data and projected that another 410 children with elevated blood lead levels might be identified per year citywide, mostly in vulnerable areas, by expanding testing in neighborhoods that tend to have higher case rates.

The highest-risk neighborhoods are in Brooklyn, Queens, and the north shore of Staten Island, and average about 12 cases per 1,000 tests, compared to less than four in low-risk neighborhoods, Afane said.

The city helps coordinate care for children with elevated levels and also works to reduce lead hazards. Since 2005, the number of New York City children under 6 years old with elevated blood lead levels has dropped 93%, a city report says.

Using a Data-Informed Strategy

But the study recommends a better, data-informed, strategy to focus more lead testing on high-need areas. “What we wanted to highlight here is that this needs to be done and reported at the neighborhood level, not at the city level,” Afane said.

The study also recommends awareness campaigns in high-risk areas emphasizing early detection, and it calls on local authorities to step up monitoring of water quality and blood lead levels in pregnant women.

“Our main goal was to use data science and machine learning tools to genuinely improve the city,” Afane said. “Data analysis is a powerful skill that could be used much more often to make a positive impact in our communities.”

Working with one of her Fordham professors at the intersection of machine learning and neuroscience, Rabia Gondur devised an innovative way to understand how an insect’s brain functions during natural movements.

When you do something simple like pick up your phone or wash your hands, what’s happening in your brain? Quite a lot, actually—neurons are firing everywhere because of all your minor movements, not to mention background activities like respiration.

“Your brain is not just stopping to do this one activity,” said Rabia Gondur, FCLC ’22, a computational research scientist at Cold Spring Harbor Laboratory on Long Island. “It’s very noisy in the brain.”

Cutting through this noise to see which movements fire which neurons is the subject of her research, which she’ll soon present at a prestigious international conference on machine learning.

Gondur devised an innovative approach with help from one of her professors, Stephen Keeley, Ph.D.—a collaboration that began easily during her senior year when his presentation in one of her capstone courses spoke to her interest in research. “I just reached out to him, and he was super accommodating,” she said.

They worked on the research while Gondur—an integrative neuroscience major—completed the requirements for the accelerated master’s degree program in data science at Fordham’s Graduate School of Arts and Sciences, after which she landed her job at Cold Spring, where she is part of a computational neuroscience research group.

She will present her research at one of the world’s leading forums for machine learning, the annual International Conference on Learning Representations, taking place in Vienna, Austria, in May.

Using Machine Learning to Study Day-to-Day Brain Function

Gondur’s research is one of many studies seeking to understand a brain’s response during complex, natural behaviors, building on prior studies of more basic movements—for instance, what happens in a monkey’s brain when it reaches left versus right in response to a prompt.

The eventual goal is to get beyond laboratory studies to see, in detail, how the human brain naturally functions. “That’s ultimately what neuroscientists are interested in understanding, is how the brain works in our day-to-day lives,” said Keeley, an assistant professor of natural science who runs a machine learning lab on the Lincoln Center campus.

But to work toward this goal, scientists have to start small—literally. For their study, Keeley and Gondur examined the brains of insects: a fly grooming itself and a moth flitting around to follow a moving image of a flower. For this, they relied on data that their collaborators at other universities gathered using brain imaging technology.

Keeley and Gondur devised a machine learning algorithm to find links between the bugs’ brain signals and the subtleties of their movements, as captured in video stills. It differs from similar algorithms because they added processes to make the measurements more precise and the results easier to interpret.

A New Tool for Brain Research

Such techniques could one day illuminate everything from brain-based diseases to variances in people’s motor skills, Keeley said. For now, their model gives a new tool to scientists trying to tease out relationships hidden in complex data. “If you are interested in genomics, if you’re interested in medicine, if you’re interested in just anything, you can basically tweak the model,” Gondur said.

Keeley is always working with undergraduates on research projects tailored to their skill level. “Rabia came in with quite a good amount of talent, and so I was able to give her a very challenging project, and she was very successful,” he said.

Their method for introducing machine learning in chemistry classes has been honored with the inaugural James C. McGroddy Award for Innovation in Education, named for a donor who funded the award’s cash prize. (See related story.)

The recipients are Elizabeth Thrall, Ph.D., assistant professor of chemistry; Yijun Zhao, Ph.D., assistant professor of computer and information science; and Joshua Schrier, Ph.D., the Kim B. and Stephen E. Bepler Chair in Chemistry. They will share the $10,000 prize, awarded in April.

Chemistry and Computation Come Together

The three awardees’ project shows how to reduce the barriers to learning about programming and computation by integrating them into chemistry lessons. The project came together during the COVID pandemic—since chemistry students were working from their computers, far from the labs on campus, it made sense to give them some computational projects, in addition to experiments they could conduct at home, Thrall said.

Because little had been published about teaching machine learning to chemistry students, she got together with Schrier and Zhao to design an activity. Zhao, director of the Master of Science in Data Science program at Fordham, involved a student in the program, Seung Eun Lee, GSAS ’22, who had studied chemistry as an undergraduate.

Their first classroom project—published in the Journal of Chemical Education in 2021—involves vibrational spectroscopy, used to identify the chemical properties of something by shining a light on it and recording which wavelengths it absorbs. Students built models that analyzed the resulting data and “learned” the features of different molecular structures, automating a process that they had learned in an earlier course.

For another project, the professors taught students about machine-learning tools for identifying possible hypotheses about collections of molecules. Machine learning lets the students winnow down the molecular data and, in Schrier’s words, “make that big haystack into a smaller haystack” that is easier for a scientist to manage. The professors designed the project with help from Fernando Martinez, GSAS ’23, and Thomas Egg, FCRH ’23, and Thrall presented it at an American Chemical Society meeting in the spring.

Thumbs-Up from Students

How did students react to the machine learning lessons? According to a survey following the first project, 63% enjoyed applying machine learning, and 74% wanted to learn more about it.

“I think that students recognize that these are useful skills … that are only going to become more important throughout their lives,” Thrall said. Schrier noted that students have helped develop additional machine learning exercises in chemistry over the past two years.

Machine Learning in Education and Medicine

Zhao noted the growing applications of machine learning and data science. She has applied them to other fields through collaborations with Fordham’s Graduate School of Education and the medical schools at New York University and Harvard, among other entities.

The McGroddy Award came as a surprise. “I don’t think that we expected to win,” Schrier said, “just because there’s so many other excellent pedagogical innovations throughout Fordham.”

Eva Badowska, Ph.D., dean of the Faculty of Arts and Sciences at the time the award was granted, said the professors’ “path-breaking interdisciplinary work has transformed lab courses in chemistry.”

There were 20 nominations, and faculty members reviewing them “were humbled by the creativity, innovation, and generative energy of the faculty’s pedagogical work,” she said.

In addition to the McGroddy Award, the Office of the Dean of Faculty of Arts and Sciences is providing two $1,000 honorable mention prizes recognizing the pedagogy of Samir Haddad, Ph.D., and Stephen Holler, Ph.D., associate professors of philosophy and physics, respectively.

“We hope to create a platform that will allow scholars and the general public to access data across museums through a simple and visually appealing online interface,” said Laura Auricchio, Ph.D., dean of Fordham College at Lincoln Center, a co-principal investigator for the project.

Several representatives from major museums and libraries, including the Metropolitan Museum of Art, the Museum of Modern Art, and the Library of Congress, were present at an October project workshop at Fordham. Joining them were scholars from Fordham, Harvard University, MIT, the New School, Sciences Po of Paris, and University of Potsdam in Germany. The group has been collaborating continually to produce a final report for the NEH in March, after which they’ll seek additional funding for the project.

Connecting Museums and Their Data

Auricchio said that the project is similar to how museums are connected in the physical realm through the exchange of traveling works of art, but instead of art they would be exchanging research data, or metadata, spawned by their collections. Auricchio distinguished the two data sets by using museum “tombstones” as an example. Tombstones are the placards one sees beside a painting in a museum. The metadata would be the boldfaced information found at the top of the placard: the name of the artists, the years the artist lived, the name of the work of art, and the medium. The research data would be the paragraph below the metadata, which would include more nuanced and detailed information about the painting: its history, influences, and place within art history. Also included in the research data would be essays from exhibition catalogs.

“Only a fraction of a museum’s holdings are photographed for catalogs, the rest is represented through this research data and metadata,” she said.

This new platform would help foster “a new kind of knowledge production for scholars, artists, curators, educators, and an interested public,” she said.

Anne Luther, Ph.D., a co-principal investigator on the project, said that one of the primary challenges is that museums publish their data in silos, and even within institutions the internal databases don’t necessarily follow the same protocol. Luther, along with Auricchio, brought the NEH-funded project to Fordham.

“A museum may have one database system they are using, but from department to department they are using it differently,” Luther said at the October workshop. “The goal is to make this data available as a public good, but at the moment they’re [the data]  not speaking to each other.”

The challenge in dealing with large institutions is that the computer science protocols have already been established, in many cases over the course of years. Luther said there have been long-standing efforts that try to connect museum data internationally, but projects that have tried to impose new standards and new protocols have failed.

“We’re not trying to bring new standards to describing metadata, but rather we want to build, on one side, a protocol that would allow us to connect them,” she said. “We want to allow for the diversity of metadata on object descriptions within the museums to remain the same. We’re not asking the museum to rewrite. We’ll fish that out.”

Speaking the Same Language

Of course, “fishing” for common phases that describe a period, or a work of art, is also one of the great challenges for the project.

Sarah Schwettmann, a graduate student at Massachusetts Institute of Technology’s Center for Brains, Minds, and Machines, said a protocol layer that aligns metadata from museums’ digital collections could be the best route.  She noted that with machine learning, which is akin to artificial intelligence, there are increasingly more tools that allow computer scientists to work with and analyze metadata. She said the resulting platform needn’t be a simple search engine or website, but could be something more.

“We could build a protocol that actually asks, ‘Can we compare how different museums talk about items in their collection?’” Schwettmann said at the workshop. “This interface would allow one to interoperate specific terms and cultural language that the various museums have developed over time. This is important because each museum develops bodies of scholarship that are specific to that institution.”

“We want a protocol layer that points back to how individual museums talk about their objects and allows users to interact with and see the diversity in terminology,” she said.

One-Stop Research

Matthew Battles, associate director, metaLAB at Harvard University, noted that today art historians will often need to travel from several galleries, museums, and archives in order to gather the strands of a story about a particular artist, particular genre, and particular period.

“We want to facilitate the research activity of a scholar who wants to tell those stories across an institutional context so that rather than spending five years visiting 25 institutions, they could have access to the data of those various institutions in one place,” he said.

He noted that while diverse institutions feature objects from similar periods in history, they may interpret that history differently. As an example, he noted that all institutions agree there was a Byzantine era, though not all agree on a start date or end date. Where one researcher might want to have a numerically specific date, another might be interested in how various institutions have defined Byzantine.

He said that rather than proposing yet one more system to bring all of the museum systems into alignment, which hasn’t worked anyway, it would be better to provide a “roadmap” of how you can bring the various data into agreement or, if one chooses, eliminate the distinctions.

Battles said the NEH seed money—known as a discovery grant—was key, since the resulting research would be a public good that could impact the way stories are told at exhibitions, in elementary school classrooms, and in higher education, all of which would be “more richly informed by a broader array of resources.”