Threads Bio

Free energy methods and digital transformation of drug discovery

Free energy predictions are well established as a robust and accurate method of triaging potential drug candidates and guiding optimisation. Now the synergistic application of AI-based generative chemistry to free energy methods promises to streamline and industrialise molecule discovery and optimisation. Drug discovery has always been an extremely difficult field. The list of criteria that a successful drug must fulfil is often very long, and many of them are rooted in the complex interplay of chemistry with biology. As a result, the optimisation of an initial hit molecule into a clinical candidate has always been at least in part a numbers game: you keep making changes until you get lucky. The intuition and experience of a good medicinal chemist can certainly increase the chances of success, as can the predictions from computational chemistry models, but neither of these are close to infallible. However, the adoption of free energy perturbation calculations (FEP) over the last decade or so has been game changing. One of the key properties needed in a drug molecule, the ability to bind strongly to the target protein, is now amenable to calculation with a good degree of predictive accuracy. Every large pharma organisation has invested significant resources in running FEP calculations because they can efficiently answer the primary question in drug development: how well is this molecule going to bind to my protein target? Free energy methods involve simulating the behaviour of a molecule both bound to a protein and solvated in water which, together with a variety of thermodynamic tricks, allow the binding constant to be either calculated directly or indirectly by comparison to a related molecule whose binding is known. These methods have a long pedigree: the theoretical foundations were put in place nearly three quarters of a century ago. However, there were always two major difficulties in making use of the method. The first is our modelling of chemistry: we need to be able to compute the energy of organic molecules with a high degree of accuracy, and to do so extremely quickly. Quantum mechanics is accurate but not quick, so we rely on approximate classical methods, and these require extensive parameterisation to get sufficiently accurate answers. The second is sampling: for accurate results, you need to explore all of the different configurations that the system can take up. For a small molecule in vacuum that is not too challenging, but adding water and including a large binding partner such as a protein makes sampling much more difficult. It is only relatively recently, with the advent of fast and relatively affordable GPUs, that free energy methods have moved from specialist calculations that required significant time on a supercomputer to everyday jobs run on a handful of computers (or even a single workstation). The hard questions around druglikeness, metabolism, pharmacokinetics, toxicity, crystal polymorphs and so forth are still only loosely amenable to computational prediction, if at all. This expansion of the use of free energy methods is enabled by widening their domain of applicability. The original methods have been expanded to solve technical problems around the calculation of binding affinities on charged molecules, peptides, and macrocycles, as well as handling complex cases where multiple tautomers and/or protomers can contribute to binding. The range of acceptable targets has also expanded to include membrane-bound proteins as well as DNA/RNA binders and disordered proteins. Exciting work is taking place on alternative ways of performing free energy calculations, such as alchemical transfer methods, multistate free energy methods, nonequilibrium methods and more. Access to these methods is also being simplified with the release of well-tested open-source frameworks for these calculations, such as SIRE and the Open Free Energy consortium. In the last few years, the democratisation of free energy calculations has led to an explosion in their use in drug discovery. Every large pharma organisation has invested significant resources in running FEP calculations because they can efficiently answer the primary question in drug development: how well is this molecule going to bind to my protein target? That is not to say that small molecule drug discovery can now be run entirely computationally: the hard questions around druglikeness, metabolism, pharmacokinetics, toxicity, crystal polymorphs and so forth are still only loosely amenable to computational prediction, if at all. However, when optimising for these properties being able to rule out, with a high degree of accuracy, which modifications to the lead molecule will harm the primary binding endpoint and which are acceptable (and being able to predict selectivity against related antitargets ahead of time) can lead to startling efficiency gains in the overall hit-to-lead and lead optimisation process. This can introduce significant savings in the very expensive and time-consuming drug discovery process. Advances in available computing power have also contributed to the increasing industrialisation of the drug discovery process. Beginning in the late 1990s the focus started to shift from the careful, handcrafted synthesis of individually designed molecules towards bulk synthesis in combinatorial libraries. However, this was always limited by the inability to focus libraries effectively on the useful portion of chemistry space. Even with the inherent efficiencies of combinatorial synthesis, the cost (of both money and time) of synthesising and assaying large numbers of molecules that failed to reach the desired efficacy threshold was significant. An important factor in facilitating the widespread use of free energy methods is the ease of use of the software used to run the simulations. Mirroring the shift that was seen in synthetic approaches, the application of free energy calculations has moved in recent years beyond its starting point of carefully overseen and hand-tweaked calculations on individual molecules to a much more industrialised approach. Large numbers of candidates for synthesis are routinely triaged computationally to narrow chemical libraries down to the desirable regions of chemistry space. The overall hunt for a clinical candidate has been made significantly more efficient, with each cycle in the process consisting of the synthesis of a library of compounds pre-screened for high binding to their target. One application of free energy methods that has yet to reach its full potential is the synergy with AI-based generative chemistry. Generative chemistry offers a step change in how we approach drug discovery, with the promise that an AI can learn the underlying features behind protein-ligand binding and hence design better molecules than a human expert. Similarly, advances in AI protein-ligand co-folding models may remove the requirement for good-quality experimental structural biology data. However, there is some debate about whether current models truly do manage to learn chemistry or whether they just memorise their inputs. In the meantime, generative chemistry methods are widely used but are generally inaccurate, such that post-scoring of the generated molecules is necessary. Free energy methods would be ideal for this but, since molecules created by generative chemistry methods are usually structurally unrelated to each other, this needs to be done by much more computationally expensive and less accurate absolute binding free energy (ABFE) calculations, which do not require a congeneric series. Advances in computing power and improvements in the underlying methods will no doubt make the use of ABFE to triage and validate molecules created by generative chemistry a routine approach in the future. An important factor in facilitating the widespread use of free energy methods is the ease of use of the software used to run the simulations. When calculations took days and required a large amount of specialist knowledge to run, it was perhaps acceptable for the software to be hard to drive and very complex. However, in a world where it is expected that every candidate for synthesis should have been checked by a free energy calculation beforehand wherever possible, the requirement from users is that running the calculations should be as simple and as automated as possible. In the Flare™ package from Cresset, for example, the setup of a free energy calculation is not only fully automated, but the software checks for and warns about a large number of possible failure modes and system features that can lead to less accurate results. This reduces the likelihood that a calculation can be run that could give erroneous or misleading results. This guided approach maximises the efficiency of experienced computational and medicinal chemists, while also ensuring that less experienced or knowledgeable users can feel confident that the experiment they are undertaking will be accurate. Just as important is the software's ability to easily access computing resources in the cloud, as many organisations do not have direct internal access to the computing resources required to run free energy calculations at scale. With mainstream adoption of free energy methods across pharma and biotech, drug hunters are further pushing its boundaries and ever expanding its domain of applicability. Software that is easy to use while being powerful and scalable is helping to drive this progress.

Chemistry World

Mon, 28 Apr, 12:26 PM UTC

FutureHouse Unveils AI Platform to Accelerate Scientific Discovery in Biology

FutureHouse, an Eric Schmidt-backed nonprofit, launches a suite of AI tools aimed at revolutionizing scientific research, particularly in biology. The platform includes specialized agents for literature review, hypothesis generation, and experiment planning.

3 Sources

Fri, 2 May, 8:02 AM UTC

AI Designs Over 7,000 New Proteins to Help Speed Up Medicine and Research | Newswise

Newswise -- A research team from Chongqing University and Zhejiang University has developed an AI tool that created 7,245 new proteins -- entirely through computer design. These proteins are made to latch onto targets like viruses or cancer markers, and can be used in medicines, lab tests, and scientific research. Why Make New Proteins? Special proteins can act like tiny tools that grab onto harmful cells or help detect diseases. They're essential in making drugs, vaccines, and diagnostic kits. But creating them by hand takes time and money. That's why researchers and companies are turning to AI to make the process faster and more efficient. Faster Design for Biotech and Pharma The AI tool helps cut months of lab work down to weeks. It doesn't just make proteins -- it also predicts which ones will dissolve well in liquid and hold their shape when heated, so they're less likely to clog lab equipment or break down during storage or transport. This makes them ready for testing and easier to manufacture. "Our AI pipeline lets us design high-performance proteins quickly and at low cost," said Prof. Weiwei Xue. "This gives smaller labs and companies a chance to compete and innovate." Big Data, Fast Filtering To begin, the team collected over 1,300 existing protein structures from a public database. The AI generated five new versions of each, then quickly screened them for key features like stability and ability to stick to their target. Only the best-performing ones were selected -- saving labs time by skipping weak candidates. Ready for Real Use in the Lab and Clinic More than 70% of the new proteins were predicted to dissolve cleanly in liquids like water or buffer solutions -- important for use in lab tests, injection drugs, or test strips. About 60% stayed stable at high temperatures, which helps during shipping, storage, or sterilization. Designed for Flexibility and Fit The proteins were based on 55 different structure types, including parts of antibodies already used in approved drugs. The variety helps ensure the proteins can fit many different types of disease targets. Nearly half were predicted to bind even more tightly than older designs -- important for accuracy in tests and strength in treatments. Why It Matters This work shows how AI can help speed up and lower the cost of developing proteins used in medicine. For drug and biotech companies, that means a faster pipeline. For hospitals and labs, it could lead to more reliable and affordable tests. And for researchers, it's a new way to explore biology -- faster and smarter than before.

newswise

Thu, 5 Jun, 6:16 AM UTC

Q&A: A generative AI technique for designing RNA with improved function

Ribonucleic acid, also called RNA, is a molecule present in all living cells. It plays a critical role in transmitting genetic instructions from DNA and creating proteins. With the power to execute a plethora of functions, the little RNA "messenger" has led to important innovations across therapeutics, diagnostics, and vaccines, and made us rethink our understanding of life itself. A team of researchers from Boston University's Biological Design Center and the Department of Biomedical Engineering recently made significant steps forward in the development of the next generation of computational RNA tools. They recently published a study in Nature Communications describing a generative AI technique for designing different types of RNA molecules with improved function. Much like a large language model that can be used to compose entirely new texts, the model can compose new RNA sequences tailored for specific tasks in the cell or in a diagnostic assay. Their research has shown that it's possible to predict and generate RNA sequences that have specific functions across a broad array of potential applications. In this Q&A, Dr. Alex Green, Associate Professor of Biomedical Engineering at BU, discusses the power of sequence and structure of RNA molecules (SANDSTORM) and Generative Adversarial RNA Design Networks (GARDN) as new tools in developing diagnostic and therapeutic RNAs with improved functions. Why are you interested in engineering RNA with novel functions? RNA engineering has tremendous potential because RNA is the only system in the body that both encodes genetic information and has diverse functional activity across a broad array of cellular processes. Engineered RNAs have applications in gene editing for research or therapeutic purposes, diagnostics, and synthetic biological systems. Additionally, RNA engineering could provide a simpler and more streamlined approach to ensure cells produce the right amount of protein at the right time, which has tremendous potential in enabling novel protein-based therapeutics and targeted therapies. The Green lab focuses on engineering RNA in a programmable way so it can function as a device to achieve a specific end goal. For instance, we work on designing RNAs that are active in cancer cells to produce a therapy, but are completely silent if they encounter a healthy cell, which minimizes unwanted side effects. What have been the challenges to date in realizing the potential of RNA engineering? A key challenge has been the need to synthesize and screen RNA in experimental systems, which can be both time-consuming and cost/resource intensive. Several computational tools have been developed to overcome the challenges of experimental approaches, but they all use different coding platforms and architectures, which makes it very difficult to integrate them. Additionally, most of these existing methods have been designed to predict the function of specific types of RNAs, which means there isn't a single tool that can be broadly applied to answer all the questions and make all the predictions we'd like to make. How do SANDSTORM and GARDN overcome these challenges? SANDSTORM is a deep ML approach that incorporates information about RNA sequence and RNA secondary structure to predict the function of diverse classes of RNAs. We can use SANDSTORM neural networks -- which learn and improve over time as we get more data -- to predict the functional activity of the ends of RNA molecules (which play important roles in RNA stability, trafficking, and translation), parts of the RNA that interact with ribosomes, and RNAs used in CRISPR diagnostics. GARDN is a generative adversarial network architecture, which is a system tasked with generating realistic examples of functional RNA and discriminating between realistic and unrealistic examples. When we combine SANDSTORM with GARDN, we have a powerful system that can generate and select RNA sequences that provide desired functions while also being highly computationally efficient. Fewer parameters are needed for training and prediction compared with other computational approaches, which makes the system faster and related workflows easier. A graduate student can use this system on their personal computer, as Aidan Riley, a Ph.D. student in my lab who championed the work, can attest. In fact, Aidan was a driver for this project because he wanted to use ML to screen and test engineered RNA faster and more efficiently, and his background in ML played a critical role in the ultimate design of the system. What are the next steps in this research? To date we have demonstrated the utility of SANDSTORM and GARDN in engineering the ends of RNA molecules (known as the 5-prime and 3-prime ends). Part of our current focus is on designing the coding region between the two ends. Toward this end, we are merging these computational tools with other developments at BU, including self-amplifying RNA technology and more specific delivery of therapeutics. Self-amplifying RNA technology is a more efficient way to generate the RNAs we want to evaluate in experimental systems. Further down the road, we also want to engineer RNA to improve the efficiency of protein production, which could have important implications in enabling new targeted therapies and making the production of therapeutic proteins more efficient. In what ways might this research help further the development of vaccines or drug therapies, or in the treating of diseases? This research can help make vaccines and therapies smarter and more effective. Using our models, we are finding that we can significantly increase the amount of payload protein that an mRNA generates. This is valuable since it can lead to more potent treatments or it can be used to reduce the amount of RNA that is required for the therapy, lowering costs and potentially reducing side effects. We also use the models to design RNAs that only activate in response to different biomarkers that they detect. This capability could be used to develop therapies that only activate in specific tissues or in tumor cells only, for example. How could biotech companies potentially benefit from the knowledge gained by this research? Our technology could be used to help biotech companies develop RNA drug candidates much faster and more efficiently. Training new machine learning models often requires testing tens of thousands of different candidates, which is expensive and time-consuming. With SANDSTORM, we've found that we can generate effective models from only a few hundred test sequences, which is considerably smaller than I thought would be possible. For small biotech companies in particular, this cost and time benefit could be decisive in getting a new drug to market.

Phys.org

Thu, 15 May, 12:20 AM UTC

AI can transform innovation in materials design - here's how

By leveraging both domain expertise and data-driven insights, AI-enabled platforms are poised to transform materials science. Picture this scenario: David, a materials scientist at a top artificial intelligence (AI)-enabled lab, collaborates with an AI platform to design a new high-performance composite. The platform instantly generates millions of unprecedented molecular structures, screens their feasibility, predicts their properties and proposes cost-effective synthesis pathways. It then autonomously generates synthesis tasks and dispatches them to high-throughput experimental equipment. Through iterative cycles informed by rapid AI feedback, David achieves a breakthrough formula. The system evaluates his findings, suggests further refinements and alerts the project manager to begin scale-up production. This scenario is likely to be a common day for high-tech companies developing new materials in the near future. Thanks to advances in AI, material scientists today are making breakthroughs at unprecedented rates, addressing the urgent need for novel materials that enable decarbonization across all sectors. Traditional materials often fall short for next-generation green technologies, creating a bottleneck in climate response efforts. However, as highlighted in the World Economic Forum's report Top 10 Emerging Technologies of 2024, AI is revolutionizing how we discover and apply new knowledge, potentially unlocking the advanced materials required for more efficient solar cells, higher-capacity batteries and critical carbon capture technologies - accelerating our path to carbon neutrality. This technological transformation comes at a critical juncture. As global competition for finite resources intensifies, and the urgency to address climate challenges grows, breakthroughs in advanced materials will be foundational for delivering cleaner, more resilient technologies. Traditional approaches to materials development - primarily driven by experience, intuition and trial-and-error methodologies - are increasingly insufficient to meet current challenges. Materials scientists and engineers today face several critical hurdles: This is precisely where 'AI for science', or AI4S, offers transformative potential. Advanced machine learning models trained on extensive datasets, combined with high-throughput computational methods, can predict properties for numerous substances in minimal time, rapidly screening candidate materials against desired parameters. Beyond discriminative models that predict material properties, generative models can directly design molecular structures based on desired properties, creating the molecules needed. Deep Principle's latest generative model, ReactGen, for example, can propose novel and complex chemical reaction pathways by learning underlying reaction principles, enabling efficient and innovative synthesis route discovery. These models can also generate and iteratively explore diverse synthesis pathways, incorporating chemical information and physical constraints to recommend feasible reaction routes. In product formulation, AI enables multi-objective optimization to meet complex and varying market requirements precisely, saving significant human capital, material resources and development time. Global technology leaders, from Microsoft and Google to Lawrence Berkeley National Laboratory, have launched bold initiatives such as MatterGen and GNOME, using AI to vastly augment the scale and precision of materials research. In parallel, emerging players like XtalPi are constructing integrated data-generation and inference "flywheels" - combining automated laboratories with advanced AI - to enable accelerated, iterative discovery at unprecedented scale. "2024 has also been a transformative year for startups in the AI for science ecosystem, particularly in biotechnology and the emerging fields of chemistry and materials science," Chenru Duan and other AI for science community organizers wrote in their annual review. In biotech and pharmaceuticals, several major milestones have been achieved. XtalPi went public with a valuation of $2.5 billion, and other notable companies, such as Terray Therapeutics and Iambic Therapeutics, closed significant fundraising rounds while advancing their drug discovery pipelines. A standout case was Isomorphic Labs, a Google DeepMind spinoff. The company announced strategic partnerships with Eli Lilly and Novartis, securing $82.5 million upfront and projecting a potential total of $3 billion (excluding royalties). This partnership underscores the immense business value of specialized AI models, even for well-established pharmaceutical giants. Meanwhile, startups focused on chemistry and materials science have begun to emerge. Companies like Orbital Materials and DP Technology released large pre-trained machine learning potentials - Orb and DPA-2, respectively - designed to accelerate molecular dynamic simulations with higher precision. Additionally, new startups such as CuspAI, Lila Sciences and my own, Deep Principle, have successfully closed seed funding rounds, each with a vision of reshaping chemistry and materials discovery. In the case of Deep Principle, we are transforming material innovation from discovery to design by using generative AI models. We integrate generative AI, quantum chemistry and automated experimentation into a unified workflow to achieve the full chain from molecule generation and synthesis design to reaction and formulation optimization, where our self-developed generative AI platform will be one of the keys. Despite the promise of this new paradigm, significant challenges remain before AI4S can achieve widespread industrial implementation: To overcome these obstacles, businesses and research institutions tend to forge collaborative partnerships to create universal or domain-specific datasets, while continuously improving algorithms through iterative development cycles. Crucially, integrating purely data-driven approaches with first principles methodologies enhances extensibility, while deep learning algorithms excel at fitting available data, first principles and domain knowledge can effectively extrapolate to areas with limited or no empirical data. AI is rapidly becoming a pivotal force, ushering in a new era of materials innovation. Its unrivalled ability to scan and assess vast landscapes of known structures and combinations empowers scientists to quickly identify optimal materials that would otherwise take years or even decades to discover by traditional means. Generative AI models go a step further, directly creating new-to-nature molecules and reaction pathways tailored for specific applications, unlocking unprecedented possibilities in reverse-design and innovation. When integrated with automated experimental platforms, AI enables a fast, efficient cycle of hypothesis, prediction and validation, drastically accelerating the progression from theoretical discovery to scalable industrial production. Far more than just added convenience, this agility directly addresses longstanding bottlenecks inherent in conventional materials R&D, which is often time-consuming, labour-intensive and marked by low success rates - falling short of what fast-moving markets and industries increasingly demand. By fostering seamless interdisciplinary and even cross-border collaboration, and by leveraging both the depth of domain expertise and the breadth of data-driven insights, AI-enabled platforms are poised to transform materials science - delivering the breakthroughs necessary for a more sustainable and robust future for all.

World Economic Forum

Tue, 24 Jun, 3:26 PM UTC

Threads Bio

Free

About Threads Bio

Related fields

Related News

Similar products

Meta Threads Analytics

AutoThread AI

BenevolentAI

SynthaceGPT

Jadbio