The efficiency of bringing new drugs to the market is amazing: About 90% of new drugs fail in clinical trials, with development time of 10-15 years, and the cost may exceed $2 billion. It’s hard to think that more efforts require improvements in AI, and the technology industry (all of the recent advancements are exciting) is diving into it.
But, what will make us here?
History tells us that the right time and the right equation can change everything. Einstein's E = MC2 Help usher in the nuclear era. Neural networks have sufficient computing power and training data to ignite the current AI explosion. In the late 1990s, when it was difficult to find anything online, Sergey Brin and Larry Page invented the Pagerank algorithm to make Google (now Alphabet) one of the most valuable companies in the world.
Pagerank and other so-called “central algorithms” may not have changed the world yet. In fact, they could be the key to the next AI-driven breakthrough in drug discovery.
When applied to a website, the central algorithm determines which pages are most linked to which page, and therefore is most relevant to the query. When applied to biomedical data, they can identify the most linked answers about scientific questions, highlighting which findings have the strongest experimental support. Crucially, central algorithms can be applied to relatively raw data, including large sets of data generated by modern high-throughput methods, so they can connect points that have never been connected before, connecting between data points spanning countless databases and other data sources. New connections can mean new discoveries. Moreover, multi-proxy AI systems have more revolutionary features than in the past.
A lot of data, too few explanations
According to the design, the scientific publication tells stories, and only a few stories fit every paper. Thus, modern research and its accompanying massive datasets leave thousands or even millions of stories. When used in conjunction with other research, the number of myriad stories increases, perhaps exponentially.
It was immediately a tragedy and a huge opportunity. Some of these stories may be new strategies to cure cancer or rare diseases, or to deal with important public health threats. We lack them simply because we cannot use data that is already in the virtual hand.
The calculation of fast echoes makes us talk about how much data: a 2022 survey found about 6,000 publicly available biological databases. One of these databases, the public repository hosted by NCBI, currently has nearly 8 million samples. If we assume that each sample has about 10,000 measurements (half of about 20,000 genes in the human genome), about 80 billion measurements will be obtained. Multiplying by 6,000 databases brings us to about 500 trillion data points. That's not a computational chemistry database, a proprietary data source or a large-scale data set No Stored in a central database. Whether it's real numbers, it's undoubtedly big and is growing rapidly.
Chance
Effective utilization of this treasure trove of data can significantly improve the ability of AI approaches to achieve meaningful biomedical advances. For example, by combining centralized algorithms with a construct called a “focus map”, AI agents can indeed use this data to pass experimentally supported discoveries from traceable sources. Furthermore, when used in conjunction with large language models (LLM), such as Openai's Chatgpt or Anthropic's Claude, graph-based approaches can run autonomously, resulting in profound insights into the driving forces of disease and potentially reveal new ways to treat them. At this breathtaking AI advancement, people's expectations of “troubles” are the top of “troubles”. Such a statement is understandable, but it is almost certainly premature. In fact, we may be on the eve of the next breakthrough: a new combination of “old” algorithms is expected to fundamentally accelerate the discovery and development of new drugs. Such advancements are urgently needed and leveraging the full breadth of available tools and data that may eventually be touched.
Photo: MF3D, Getty Images
As an early pioneer in microarray technology, Doug Selinger has written some of the first publications describing experimental and computational methods for large-scale transcriptional analysis. After completing his PhD in the lab at George Church at Harvard University, he joined the Novartis Biomedical Research Institute, where his 14-year career spans across the entire drug discovery pipeline, including important work in target ID/verification, high throughput screening and temporary safety.
In 2017, Doug founded PLEX Research to develop a new form of AI based on search engine algorithms. PLEX's unique platform helps dozens of biotechnology and pharmaceutical companies accelerate their drug discovery pipeline by providing interpretable and feasible analyses of large-scale chemical biology and OMICS datasets.
This article passed Mixed Influencer program. Anyone can post a view on MedCity News' healthcare business and innovation through MedCity Remacence. Click here to learn how.
