AI agents enter the fray: How can new drug development bid farewell to step-by-step problem-solving and break through cognitive blind spots? | Frontline
What changes will AI bring when it is applied to the entire process of new drug R & D?
"For AstraZeneca, AI is no longer just a single tool for drug discovery. We are building an 'AI First' R & D culture. From early - stage drug discovery to later - stage clinical translation, we hope that the entire R & D process can be deeply empowered by AI." On March 21st, He Jing, the Global Senior Vice - President of AstraZeneca and the Head of Global R & D in China, said in a media interview.
On March 20th, AstraZeneca signed a school - level scientific research cooperation agreement with Tsinghua University and announced the establishment of the "Tsinghua University (Institute for AI Industry Research) - AstraZeneca Joint Research Center for AI - Driven Drug R & D". It is reported that this joint research center is the first key implementation project under the strategic cooperation framework of both parties.
In the past few years, AI has shown great potential in AI - driven drug discovery (AIDD). However, previous implementation practices have been more concentrated on molecular generation and structure optimization. Next, using AI to discover and validate new targets, accurately analyze disease mechanisms, improve the success rate of clinical trials, and reduce the scientific uncertainty of new drug R & D have become important concerns in the current industry.
"What AI brings is not only technological iteration but also the reshaping of the paradigm of scientific discovery." Professor Lan Yanyan from the Institute for AI Industry Research at Tsinghua University (hereinafter referred to as Tsinghua AIR) pointed out that AI is bringing three progressive and substantial changes to drug R & D.
Firstly, the most direct change is the leap in efficiency. Taking January this year as an example, the Tsinghua AIR team published the ultra - high - throughput virtual drug screening platform DrugCLIP in "Science". Its screening speed has increased by a million times compared with traditional drug methods, and it has achieved virtual drug screening covering the scale of the human genome. This enables humans to explore the vast chemical space of up to 10^60 in a more systematic way. After AI quickly completes massive initial screening, researchers can focus their resources on the R & D of candidate molecules in the database after fine - screening.
Secondly, there is a change in the path of scientific discovery, from the "hypothesis - driven" approach of human scientists to the "data - driven" approach of AI. In the traditional model, R & D highly depends on top - level scientists putting forward hypotheses within the existing knowledge framework and then spending several years to verify them, which is inevitably limited by "cognitive blind spots". The new data - driven paradigm can integrate multi - modal massive data such as literature, knowledge graphs, and results of wet and dry experiments, and automatically mine and put forward underlying and fundamental hypotheses. The ability of AI to discover and understand the relationships between complex things is bringing new opportunities to highly complex tasks such as new drug R & D.
At the third level, "Now the development of AI Agents is pushing the transformation of the scientific research paradigm to the extreme." Traditional drug R & D often progresses linearly. In the process, it is difficult for researchers to foresee in advance what kind of chain reaction the current single - point decision will have in the future, which directly leads to the accumulation of errors and high failure rate in new drug R & D.
"Agents can highly integrate various AI modules and cooperate deeply with humans." Lan Yanyan pointed out. "More importantly, when making decisions, it can incorporate front - end evidence collection, mid - end hypothesis generation, and back - end future result simulation into the current decision - making closed - loop. It is not solving single - step problems but making global systematic decisions. For the extremely complex life science system, the introduction of Agents will bring qualitative changes."
The implementation of "AI scientists" in clinical development is an inevitable trend. However, there are still a large number of technical difficulties to be overcome, such as how to endow the model with long - range logical reasoning ability and how to make precise plans for extremely complex tasks. Solving these technical difficulties and transforming them into considerable value cannot be achieved without real industrial scenarios and high - quality data.
"AstraZeneca has a global pharmaceutical R & D platform and clinical translation capabilities, while Tsinghua AIR has top - notch AI algorithm capabilities and talent advantages. This combination can enable the cutting - edge scientific and technological platform originating from China to meet global R & D standards and real needs at the 'starting line', accelerate the transformation from scientific discovery to drug R & D, and enable more Chinese scientific research achievements to enter the global R & D pipeline faster and be transformed into developable innovation projects," He Jing pointed out.
It is understood that this Joint Research Center for AI - Driven Drug R & D will focus on AI - driven molecular research, translational medicine, and clinical development innovation, and promote the research results to be accelerated towards clinical application.