The field of drug discovery is facing an evolution as the pharmaceutical industry increasingly turns to artificial intelligence (AI) to curb soaring costs and reduce the time required to bring new medications to market. With the expenses associated with developing new drugs doubling approximately every nine years, a trend known as Eroom’s Law, the pressure to gain a competitive edge has intensified. On average, it now takes about 10 to 15 years and can cost between $2 billion to $3 billion to develop a new drug, with failure rates often exceeding 90%. In response, companies are leveraging AI to enhance success rates and streamline the development process.
One of the most significant applications of AI in this domain is in hit identification, where algorithms screen vast libraries of molecular entities against disease-related targets. This approach has shifted the paradigm from traditional empirical screening methods to predictive design, enabling researchers to create drug candidates from scratch and forecast their interactions with various targets. Paul Belcher, a director at Cytiva, notes that this transition allows drug developers to bypass physical limitations in screening, ultimately reducing the time and resources expended on low-quality candidates. However, AI still faces challenges, particularly in reliably predicting the kinetics and developability of new compounds, necessitating laboratory validation of AI-generated candidates.
As AI continues to drive demand for rich datasets, it also underscores the critical need for comprehensive and high-quality data. Many AI models have been trained on publicly available datasets, but as Belcher points out, these often lead to diminishing returns due to their lack of diversity and bias. Furthermore, the publication bias towards positive results exacerbates the issue, as many studies do not disclose failures. The absence of negative data impedes the model’s ability to learn from mistakes, creating a gap in understanding that could improve AI predictions. To address these challenges, there is a growing interest in tools to verify data integrity, such as Cytiva’s Image Integrity Checker, which utilizes secure hashing algorithms to detect manipulated images. Looking ahead, the vision of fully autonomous labs that operate continuously, integrating AI to guide research and optimize outcomes, could revolutionize drug discovery, making it more efficient and effective.
Source: Closing the data loop in AI-driven drug discovery via MIT Technology Review