Scientists at St. Jude Children's Research Hospital have successfully utilized artificial intelligence combined with genetic data to identify IRS4 as a promising, low-toxicity drug target for various solid tumors. The findings, published in Science Advances, establish a proof of concept for evaluating potential drug toxicity at the earliest stages of discovery, a shift aimed at addressing the high failure rates in cancer drug development. Traditionally, researchers prioritize the efficacy of new drug candidates, often leaving safety assessments until late in the clinical process. This approach contributes to a staggering failure rate, with an estimated 85% to 97% of therapeutics entering Phase 1 clinical trials failing to secure FDA approval, largely due to toxicity in normal tissues. This is particularly critical for pediatric oncology, where treatment-related side effects can result in chronic health issues decades after the initial cure. The St. Jude team, led by corresponding author Samuel Brady, PhD, sought to correct this imbalance by predicting toxicity upfront. The researchers developed a multistep screening strategy to find targets with a high therapeutic index, meaning they are effective against cancer while posing minimal risk to healthy cells. They began by analyzing genes in the Dependency Map portal, a database listing genes essential for cancer cell survival. From this pool, they narrowed the list to 346 potential targets that resembled existing FDA-approved cancer therapies. The team then employed AI to scan scientific literature for evidence of natural mutations in humans and mice where the loss of these genes resulted in no significant harm or only manageable health disruptions. This AI-driven filtering reduced the candidate list to 25 genes, a subset that included known targets as well as previously unexplored possibilities. Among the remaining candidates, IRS4 emerged as the most promising prospect. The protein was found to be rarely expressed in normal adult tissues, and individuals naturally lacking the gene are generally healthy, with only minor thyroid-related effects. Furthermore, the specific region of the IRS4 protein containing a potential drug-binding pocket is not required for its cancer-promoting functions, suggesting that targeted protein degradation could be a viable therapeutic strategy. In experiments, removing or degrading the IRS4 protein halted the growth of cancer cells that relied on it. Brady described IRS4 as an on-off switch for cancer cells, noting its presence is a strong indicator of tumor vulnerability. The study demonstrated that IRS4 dependence exists across multiple tumor types, including pediatric malignant rhabdoid tumors, osteosarcomas, and certain brain tumors, as well as adult cancers such as breast, lung, uterine, and stomach cancers. This broad applicability makes IRS4 a significant target for future drug development. The broader implication of this work is the validation of integrating toxicity prediction into the initial phases of drug discovery. By focusing on targets like IRS4 that carry a lower risk of side effects from the outset, the team aims to develop therapies that not only cure cancer but also ensure survivors can live full, healthy lives without the burden of lifelong treatment complications. This approach marks a strategic evolution in oncology, balancing the pursuit of effective treatments with the urgent need for patient safety.