A mathematical model developed by researchers in the informatics program at Children's Hospital Boston can predict a drug's teratogenicity, or risk of causing fetal abnormalities, by examining its gene targets. According to the team, the method could be a useful tool to screen new drugs in the preclinical phase of drug development.
As reported in a paper published in the March issue of Reproductive Toxicology, Isaac Kohane and Asher Schachter analyzed data on drug-target-gene relationships for 619 drugs with known pregnancy risk classifications to determine whether developmentally related functions were over-represented within the target protein families of teratogenic drugs.
Currently drugs are grouped into five classes based on their risk to fetuses: Class A drugs are considered safe, Class B drugs show little evidence of fetal risk, drugs in Class C have some evidence of risk based on animal data, Class D drugs have evidence of human risk but the benefits may outweigh the risks in some cases, and Class X drugs are known to cause abnormalities.
The analysis showed that drugs whose targets have a high proportion of genes associated with fetal development tended to be in the higher risk classes. Based on this information, Kohane and Schacter were able to create a model that showed 79 percent accuracy in predicting whether a drug would be categorized as a Class A or X drug based on the proportion of developmental genes that it targeted.
Schachter, an assistant professor of medicine at Children's Hospital, explained to BioInform that the duo began by downloading target data for more than 4,000 drugs stored in the DrugBank database and formatted the data for the R programming environment.
Next, he said, they narrowed the list down to 619 drugs by selecting those that already had a pregnancy risk classification as well as known target protein family IDs. In total, they obtained 7,426 unique GenBank gene IDs for the target protein families.
The team then randomly selected drugs from Classes A and X to serve as test and training sets. For each of 10 iterations, they selected genes targeted by Class X drugs in the training set. Then, for each test set drug, the researchers calculated a Gene Risk Quotient, or GRQ, by computing the ratio of the number of the test drug's Class X training genes to the total number of genes targeted by the test drug. They repeated each set of 10 iterations 50 times in order to calculate mean and standard deviation values for the performance metrics.
After applying the model across drugs in all five classes, the researchers found "that higher ratios correlated with increasing risk class," Schachter said.
The researchers also expanded their analysis to include drugs without a known pregnancy risk class. Of 1,881 DrugBank drugs that they determined to target developmental genes, they found that 1,325 drugs, or approximately 70 percent, are of "unknown teratogen risk class," and another 355 drugs, or 19 percent, fall into Class C risk.
"Taken together, this represents 1680 … drugs that (a) target developmental genes, and (b) have unknown effects on the human fetus," the authors wrote. "The preclinical target gene signature and GRQ may therefore enable more efficient preclinical teratogenicity screening for a large number of compounds, and might serve as another tool to help focus limited resources for post-approval surveillance of teratogenicity."
For their next steps, the team aims to further validate the model before sharing predictions for specific drugs with physicians and drug developers who they said would be likely to adopt the tool.
Since it would be unethical to give pregnant women drugs as part of a clinical trial, the team plans to study the effects of the drugs on zebrafish embryos in an attempt to validate the model. However aside from the obvious differences between humans and zebrafish, there are other challenges involved that could make it difficult to move the model into the clinic.
Schachter pointed out that while "missing digits and facial anomalies" are easy to spot, some teratogenic effects are far more "subtle" and harder to detect. Because most drugs are taken relatively rarely in pregnancy or may be taken in combination with other drugs, harmful effects on fetal development tend to be rare and can go unnoticed.
"In order to validate this finding comprehensively we would have to engage physicians everywhere to be more alert about very subtle findings ... it's virtually impossible to do," he said.
Furthermore, while there is significant data on drugs at the two extremes of the risk spectrum — Classes A and X — there isn't as much information on drugs that fall into Classes B, C, and D, which pose increasing amounts of risk to fetal development.
Schachter said the team has no plans to make the model available as a software package or as a web-based tool. He noted that the data is publicly available for anyone to download and calculate the ratios themselves, obviating the need for a specific tool.
"Prescribing drugs to pregnant women is like shorting stocks; people assume the worst first," he explained. "In the vast majority of cases the obstetrician will consider the drug to be at risk unless it’s a class A and if it's in any other class they are going to weigh the benefits and the risk for that individual patient, so we did not want to create something that would interfere with that doctor-patient relationship."
Nevertheless, the researchers indicate that the approach could be useful to drug developers "in determining a new drug’s teratogenicity risk prior to first-in-human trials, since drug targets can be determined in the preclinical phase of drug development."
Furthermore, they added, the method might encourage drugmakers to "focus post-approval surveillance efforts on higher-risk drugs, with significant public health benefits to pregnant women and their fetuses."
Have topics you'd like to see covered in BioInform? Contact the editor at uthomas [at] genomeweb [.] com.