Deep Below the Alamo

A thousand feet below the streets of San Antonio, three species of blind catfish tell a story of evolution, water quality

Below the bustling streets of San Antonio, swimming among limestone channels and caves, are more than a hundred invertebrates and two blind catfish species that have called the Edwards Aquifer home for millions of years. 

Both the Widemouth (Satan eurytsomus) and Toothless Blindcat (Trogloglanis pattersoni) diverged from ancestral relatives 30 million years ago when they invaded the darkness of the developing aquifer. Researchers believe the catfish can provide insight into the overall health of the water that more than two million people consume. Research on these species and their environment has brought together traditional field work and AI tools. 

"Improved sampling of the blindcats and other diverse organisms living with them could quickly tell us a lot about the health of the aquifer.”

Dean Hendrickson, curator of Ichthyology in the College of Natural Sciences at UT Austin, travels the southwest collecting specimens and data. The Edwards Aquifer blindcats have been a focus of his work for years, starting in the part of the aquifer that extends from Texas into northern Mexico. Hendrickson and colleagues rappelled into deep caves in Mexico to collect specimens of another species, the Mexican blindcat, and eventually discovered that species in Texas, too. 

The habitats of the other two blindcats are more than a thousand feet below San Antonio. Research relies on specimens that end up on the surface. In the early 20th century, the first specimens surfaced through artesian wells, which flow upward without pumps or machinery. 

“Someone would punch a new well and blindcats would spill out and flow down streets in San Antonio,” Hendrickson said. 

Those specimens were either dead or would die shortly thereafter but reports of them piqued the interest of scientists who formally named and preserved them. Artesian wells proliferated, and it was not long before pumps further increased the amount of water, and blindcats, being extracted. 

Today, however, most of San Antonio’s wells have been plumbed directly into distribution and treatment systems, and the ability to collect specimens by placing nets over well outlets has ended. The last-collected specimen of the Widemouth Blindcat surfaced in 1984, and a few Toothless Blindcat specimens have been collected for scientific study over the last decade, all from a single, small agricultural well southwest of San Antonio. 

“I might not expect a connection that the large language model is highlighting for me so I have to go back to determine if it makes sense, and often it does.”

In short, data collection has been hamstrung. Finding the Mexican blindcat in the Edwards Aquifer, for example, confirmed a theory that the aquifer is connected to aquifers in northern Coahuila, Mexico, the only other location this species has been found. 

“When the Edwards Aquifer in San Antonio gets to a certain level,” Hendrickson explained, “it spills over into that more southern extension of the aquifer. Conversely, the spill is sometimes in the other direction, so whatever happens in either the Texas and Mexican portions of this interconnected aquifer is going to impact groundwater in the other.” 

“We still know little about the quantity and quality of the water in this region, however, improved sampling of the blindcats and other diverse organisms living with them could quickly tell us a lot about the health of the aquifer.”

Blindcat specimens have not been collected for 40 years. Only renewed sampling can disprove what appears to be its extinction. The data available for study now is qualitative and quantitative, including newspaper accounts from the early 20th century of blindcats flowing to the surface; precise geospatial coordinates of where blindcats and other aquifer species were collected; water quality samples; and text of diverse scientific publications. 

TACC Research Scientist Suzanne Pierce was looking for a project to test an analytical environment that leverages advanced computing resources and allows a user to query heterogeneous datasets using a reusable large language modeling (LLM) tool. This provides access to LLM capabilities outside commercial services like ChatGPT. Hendrickson’s work was a perfect fit. 

“This AI approach using large language models can accelerate data curation and fusion because it can be more efficient and reproducible, in some ways, than a human,” Pierce explained.

 It does not replace the need for human involvement, but the process goes faster, and can make connections that defy expected outcomes. 

“I might not expect a connection that the large language model is highlighting for me,” Pierce said, “so I have to go back to determine if it makes sense, and often it does.” 

Pierce believes this project will advance how LLM’s merge disparate data across scientific disciplines, which could lead to discoveries more quickly than traditional methods. 

“It’s remarkable,” Hendrickson elaborated. “AI is a great way to learn things quickly and open the doors to people to ask questions. That’s where I’m hoping we’re heading with this.”