Skip to main content
Premium Trial:

Request an Annual Quote

Smitten with Computational Toxicology, Pharma Calls for Better DBs to Link Structure, Toxicity


Big pharma is eyeing computational toxicology as a promising way to reduce late-stage attrition in the drug discovery pipeline, but there is still plenty of room for improvement in the field — particularly in the area of database development, according to speakers at a recent event on the subject.

At a symposium at the New York Academy of Sciences this week called "Structure-Activity Databases for Predictive Toxicology: Current Progress and Future Directions," several speakers from pharmaceutical firms discussed how they are using computational tools to assess the safety of potential compounds before they are synthesized.

Ray Kemper of Boehringer-Ingelheim, a co-organizer of the symposium, said in his opening remarks that 30 to 50 percent of preclinical attrition is due to safety issues, and that the pharma industry stands to save substantial time and money if it can find a way to weed out toxic compounds earlier in the process.

Noting that pharmaceutical firms must find a way to balance the pressure of getting new drugs to market with that of ensuring that these compounds are safe, Kemper said that computational techniques are becoming "increasingly popular" in the industry as a means for early screening of potential compounds.

Philip Bentley, global head of safety profiling and assessment at Novartis, backed up Kemper's claim. Bentley said that Novartis launched an internally developed web-based predictive toxicology system called In Silico ToxCheck two years ago, and that it's currently the second-most visited site on the company's Intranet.

"The FDA does take this quite seriously."

Bentley said that his group feels "confident enough about the predictive value [of In Silico ToxCheck] that we abandoned the idea of high-throughput mutagenicity screening." The computational approach, which generates "structural alerts" for compounds that have a similar structure to one or more compounds with known toxicity, averages around 80 percent sensitivity and 90 percent specificity, and "is as good as any high-throughput system," he said.

Bentley stressed that the computational system is only one aspect of the decision-making process for Novartis scientists, however, and noted that it is never used as the sole basis for a go/no-go decision on a compound.

Wanted: More Data

While Novartis is seeing clear benefits from its system, other speakers at the NYAS symposium were a bit more cautious in their assessment of the potential for computational toxicology. Several noted that the field is still maturing, but progress will be limited as long as it remains difficult to access sufficient data linking chemical structure to toxicity.

Nigel Green, the head of Pfizer's global toxicoinformatics center in Groton, Conn., noted that the FDA Genotoxicity database, marketed by LeadScope, only includes 251 compounds, while its Chronic/Subchronic database only includes 84 drugs. Vitic, a toxicity database from the UK-based non-profit group Lhasa, includes around 6,000 compounds, but not all of these have sufficient toxicity data, Green said.

While publicly available databases contain more compounds than commercial databases, "the chemical space and coverage in the public databases is not the same as it would be for pharmaceuticals," Green noted. May of these databases, compiled for academic research projects and not drug discovery, contain molecules with low molecular weight and "simple substitution patterns," he said.

In addition, Green noted, these databases are not integrated and contain a large amount of overlapping data, which makes navigation difficult and time-consuming.

Green cited the "lack of data to fully understand structure-activity relationships" as the primary limitation in computational toxicology.

Boehringer-Ingelheim's Kemper said that there is currently "strong demand from the structure-activity relationship community for more data." Kemper said that while there are quite a few databases available, "some are not so good for modeling quantitative structure/toxicity relationships" because they were not developed for that purpose.

FDA (and Friends) to the Rescue

More data is on the horizon, however. Daniel Benz, database manager for the US Food and Drug Administration's Informatics and Computational Safety Analysis Staff, said that the FDA is "close to a full agreement" with the US Environmental Protection Agency to release all of its toxicity data through the FDA's databases.

The two agencies began an informal collaboration in this area earlier this year, which resulted in the EPA adding information on more than 5,000 compounds to the FDA records. [BioInform 03-03-06]

Toxicology Databases
DSSTox (US Environmental Protection Agency):
FDA Toxicity Databases (Leadscope):
Mechanism-based Toxicity Database and Toxicity Database (GVK Bio):
ToxNet (National Library of Medicine:
Vitic (Lhasa):
Predictive Toxicology Software Packages
Derek for Windows (Lhasa):
Lazar (In Silico Toxicology):
MCase (MultiCase):
TopKat (Accelrys):

Benz said at the symposium that if the negotiations remain on track, the Genotoxicity database could soon contain up to 10,000 records.

"The FDA does take this quite seriously," Benz said.

The goal of the ICSAS group is to eventually collect information on toxicity and adverse events for every chemical that the FDA has reviewed. Benz noted that he and his team of four researchers have just started this project, but he said they plan to focus on human effects this fall, when they will begin compiling data on approximately 8,600 drugs from more than 1.2 million adverse event reports.

The FDA will release all its toxicity data in the ToxML format, Benz said.

Benz said that the FDA has also collaborated with Elsevier MDL on its upcoming PharmaPendium product, which includes preclinical data, human clinical trial data, and post-market surveillance reports for all drugs approved between 1992 and 2005.

Phil McHale, vice president of corporate communications and scientific affairs at Elsevier MDL, told BioInform that PharmaPendium is "technically" available now, but the company plans to launch it formally at the World Pharmaceutical Congress in Philadelphia at the end of the month.

Benz noted that as more structure-toxicity data becomes available, predictive toxicology software will continue to improve, and pharmaceutical firms will have more confidence in these methods. Truly novel compounds will always pose a problem, however, because these methods all require structural similarity to known compounds in order to predict the likelihood of toxicity.

As for which predictive methods are best, Benz said that his team has been comparing several different software packages, and has found that they all have slightly different approaches and different coverage of the chemical space. He said that his group is working on an approach to tune different packages to improve either their specificity or sensitivity, with the goal of running them all simultaneously to achieve optimal results.

"Don't decide what's best — just use all of them," he said.

— Bernadette Toner ([email protected])

Filed under

The Scan

Tens of Millions Saved

The Associated Press writes that vaccines against COVID-19 saved an estimated 20 million lives in their first year.

Supersized Bacterium

NPR reports that researchers have found and characterized a bacterium that is visible to the naked eye.

Also Subvariants

Moderna says its bivalent SARS-CoV-2 vaccine leads to a strong immune response against Omicron subvariants, the Wall Street Journal reports.

Science Papers Present Gene-Edited Mouse Models of Liver Cancer, Hürthle Cell Carcinoma Analysis

In Science this week: a collection of mouse models of primary liver cancer, and more.