Science Firsthand is a storytelling platform from Bristol Myers Squibb that celebrates the countless moments along the path from drug discovery through development – unique moments that have the potential to lead to major scientific breakthroughs that ultimately help transform the lives of patients. The third episode highlights the exponential technological progress that is changing the face of R&D and bringing new, meaningful, high-quality medicines to patients faster and more efficiently with the use of predictive tools.
Payal Sheth [00:00]: It is in our human nature to want to know the outcome of an endeavor before we begin. Are we going to be successful? Are we going to fail? Our belief and confidence that our actions are going to translate into success is what keeps us going as human beings. That is also the primary, yet elusive goal in drug discovery. How can we predict, before years of testing, what molecules are actually going to translate into medicines of the future?
Mike Ellis [00:38]: Drug discovery has been described as trying to find a needle in a haystack. That analogy falls a little bit short because we don't exactly know what the needle looks like. We understand what we ultimately want to happen therapeutically in a human, but trying to understand what molecule we should go and make, that's the needle that we're looking for. And it's within a vast opportunity space of millions of molecules. It's a bit miraculous that, that we ever discover the molecule that is safe and effective.
Mike Ellis [01:11]:
Typically what has occurred in the past is it has been a funnel-based approach of drug discovery, where it's widest at the top and gets narrower as molecules progress. And we'll start with many, many molecules. And then working our way through and narrowing as we go, to ultimately arrive at what we believe is a single molecule that should move forward into clinical development. There are many, many, many possibilities of what could be created in the lab. We have to have thick skin. We have to be persistent. And it's a multi-year process.
Payal Sheth [01:43]: Antibody generation used to be a very empirical, very labor intensive process. In fact, it would take us anywhere from 9 to 12 months to generate the first pool of antibodies. And it would further limit the number of programs that we could work on and the candidates that we could progress from a drug discovery standpoint. The longer that we have to wait for the development processes, the longer the patients have to wait for these transformative medicines. Over the past 10 to 15 years, advances in recombinant DNA technologies has allowed us to shrink our timelines for antibody generation.
Payal Sheth [02:15]: It is very exciting to be a scientist in this era. We're experiencing a convergence built on decades of advances in different scientific disciplines. Convergence of lab-based scientists, working with computational scientists, the sheer amount of data sets that we have that's accessible to us, the compute power, the AI machine learning infrastructure, and algorithms are fueling our predictive power in a way that was completely unprecedented even a few years ago.
Payal Sheth: [02:48]: We're using artificial intelligence and machine learning in the large molecule space to help us prioritize molecules for synthesis and generation that have the highest probability of success. This is removing a lot of redundant work in our organization. It's allowing us to work on more programs, and it's allowing us to accelerate existing programs. We're continuously learning from that perspective. As we generate more data sets, we're updating our models to reflect these updated data sets so that they can become better in terms of predictive outcomes.
Mike Ellis [03:19]: We have moved from a funnel where every molecule is treated the same upfront to now saying, each molecule is unique, each molecule can have its own path for decision making. And that dramatically accelerates the process. We're going in the lab, and we're making things that have a higher probability of meeting the criteria that we're setting out for. And so that's a time saving, that's an energy saving that accelerates our drug discovery process.
Payal Sheth [03:53]: At Bristol Myers Squibb, we have a long history of using novel technologies in order to accelerate our drug discovery processes and improve our chances for success. We have integrated AI, machine learning, and the human component as a part of our drug discovery fabric. We view these technologies as an extension of our labs. There's a cultural component to embracing the mindset of combining human as well as computational aspects of drug discovery. And it is these synergies that's allowing us to ensure that we are not losing sight of what we already have in terms of extensive experience in drug discovery, but also the world that we live in that has incredible computational capabilities and that fuel our predictive molecule invention aspirations.
Mike Ellis [04:41]: We are applying what we call our predict first strategy across our small molecule portfolio. We are predicting before we synthesize on the majority of the molecules that we go into the lab and create. Just a few years ago, we were predicting maybe 5% of the molecules. It's also allowed us to move from a funnel-based approach to a tailored, dynamic screening strategy. We are seeing at this point, measurable and meaningful impact to the rate of progression and the quality of progression of our programs. And that's what we're motivated by. We want to bring more medicines to more patients faster.
Mike Ellis [05:31]: Today we have the greatest predictive power at the earliest phases of testing. This is where we have the most data. I believe that as we move forward in the future, that we will improve our predictive power in the later stages of the discovery and development process, that we will be able to predict safety, that we will be able to predict developability and other aspects that today we just don't have as much data.
Payal Sheth [05:59]: Typically, drug discovery timelines have been in the scale of decades. And it has been incredibly hard to truncate those timelines. For us to be able to leverage predictive molecule invention upfront allows us to accelerate drug discovery in a way that increases our chances of progressing the best molecules.
Payal Sheth [06:16]: I've always been a dreamer as a kid. I imagine a world where we could create and design biotherapeutics molecules in silico using our sequence and structural models that are anchored around experimental data sets to get to the therapeutic much faster. And the shorter these timelines, the better the patients are positioned in having access to these transformative therapies in the future.
What you should know
|
It is human nature to want to know the outcome before beginning a new endeavor. Will we be successful? Will we fail? Our belief and confidence that our actions will translate to success is what keeps us going as human beings, says Bristol Myers Squibb’s Payal Sheth, vice president, Discovery Biotherapeutics, Lead Discovery and Optimization. How can we predict, before years or sometimes decades of testing, which molecules will translate into the medicines of the future?
Achieving an elusive goal
With the computational power of AI and machine learning, scientists have accomplished an elusive goal in drug discovery and development: more accurate predictions of success or failure, reducing the need for iterative rounds of trial and error and greatly improving R&D productivity.
Armed with high-quality datasets that fuel predictive models, researchers at Bristol Myers Squibb are equipped to make faster decisions and better prioritize potential molecules. From virtual screening and generative molecular design to mechanistic modeling and closed-loop automation, the company has an ambitious strategy to become the first truly predictive biopharmaceutical company by harnessing the power of AI/ML to increase productivity, reduce timelines and improve probability of success across its portfolio.
As one example, researchers can now predict protein structure by leveraging an algorithm called AlphaFold, changing how they discover antibodies, the building blocks of many key therapeutic modalities. Using this tool and ML-enabled multi-parameter optimization, researchers can optimize antibodies early on to eliminate protein sequences that may pose safety or manufacturing risks. Advancements like this save time and ultimately, the shorter the timelines, the faster researchers can get transformational therapeutics to patients.
Collaborative hybrid intelligence: Computational power meets human understanding
It is important to remember that advanced computational methods are tools scientists use to broaden capabilities; they do not replace human intellect, intuition and intentionality. The combination of computational power and human understanding of targets and mechanisms of action, with subsequent human-generated data, spurs the effective and efficient creation and evaluation of quality molecules. This research paradigm of human plus computation, or “collaborative hybrid intelligence,” is widely embraced at Bristol Myers Squibb and is central to how R&D teams operate.
Teams at Bristol Myers Squibb have integrated an ambitious AI and machine learning strategy into their R&D framework. This framework leverages causal human biology to prioritize promising targets, matches modality to mechanism through predictive molecule design, optimizes the path to clinical proof-of-concept by predicting patient segments and dosing, and accelerates overall decision making to ensure a speedy path for the approval of and patient access to transformational therapies.
“One key aspect of our research strategy is matching the modality to the molecular mechanism of action, or predicting the structure and function of the molecule needed to induce the desired pharmacology and ultimately the outcome for a patient,” says Bristol Myers Squibb’s Stephen Johnson, PhD, scientific vice president, Discovery and Development Sciences. “AI and machine learning tools can bolster researchers’ ability to make the correct match, providing new insights around how a target should be modulated and what modality is most likely to be successful. They also help democratize scientific knowledge and empower a focus on high-impact hypotheses.”
Maximizing the benefits of these tools in R&D decision making requires new thinking in terms of the structure of scientific teams.
“We have embedded highly specialized scientists, including computational scientists and information technology (IT) experts, into our teams as we incorporate AI and ML into the fabric of our scientific endeavors. We’ve embraced the mindset of combining human as well as computational aspects of drug discovery, not losing sight of our extensive experience but elevating it with new capabilities that help us fuel our predictive molecule invention aspirations,” says Sheth.
Embracing new tools to accelerate human curiosity
Today researchers have immense computational power at their fingertips, with programs specifically designed to solve biological problems.
Increased computational capacity can lead to arguably exponential possibilities and solve many of the challenges long experienced in drug discovery. AI and machine learning methods can go beyond pattern recognition, making sense of complex human biology and organizing corresponding data sets, thus empowering researchers to draw mechanistic insights from an abundance of human genetic and clinical data. More computational power can lead to more advanced simulations of biological processes, improved molecular optimization and the ability to do more experiments in silico (simulated in computers), ultimately increasing productivity and allowing researchers to move discovery programs faster.
Achieving these improvements requires not just computing power, but also high-quality data and industry collaborations that leverage the latest advancements. Extensive data sets generated from decades of game-changing innovation across therapeutic areas, from pre-clinical to real-world data, and generative AI capabilities extend the predictive power and efficiency of scientists.
AI in clinical trials Separate from the efforts in drug discovery centered on predictive molecule invention, AI and machine learning are also beginning to play a role in the way clinical trials are designed and conducted. These technologies could help make the clinical trial experience easier for administrators, investigators and patients; accelerate trials; analyze data; and recognize patterns and signals that may be otherwise missed. Researchers at Bristol Myers Squibb leverage the abilities of AI — including the digital twin approach — to integrate simulated data from underrepresented patient populations when diversified data may not be immediately available. |
Unlocking unprecedented research potential
If the last decade of technological advancement is any indication, we can expect exponential expansion of knowledge and capabilities in R&D in the near future.
“AI and machine learning tools aren't changing what we do at Bristol Myers Squibb, which is discover, develop and deliver transformational medicines to patients, but they are changing how we do it," says Robert Plenge, MD, PhD, executive vice president, head of Research. “These technologies are enabling our scientists to more deeply understand human biology and make more effective use of vast amounts of data. The advancements in predictive molecule invention seen to date have already been immense, informing key aspects of our research strategy as we look to improve the quality and speed of our investigational programs.”
By harnessing the immense potential of the symbiotic relationship between human ingenuity and cutting-edge technology, the research community is poised to enter a new era in biomedical science and change the lives of patients, one breakthrough at a time.
