Artificial intelligence (AI) is driving a fundamental transformation in drug development by shifting from the prediction of natural proteins to the design of entirely new ones. Advances in machine learning have enabled scientists to create de novo proteins with optimized structures, functions, and therapeutic properties that nature never evolved. AI models are compressing development timelines, improving precision, and expanding the boundaries of what is possible in biotechnology — from next-generation therapeutics to responsive biosensors and synthetic enzymes. As powerful new platforms emerge, questions of transparency, reproducibility, and regulation have become central to the field’s future. Ultimately, AI is not only accelerating drug discovery but redefining it — opening the door to programmable biology and a new era of rationally designed medicines.
Introduction – From Prediction to Creation
For decades, the pharmaceutical industry has been shaped by a fundamental reality: proteins are designed by nature, and drug development has largely meant discovering, modifying, or mimicking them. The emergence of artificial intelligence (AI) in protein science has upended that assumption. AI-driven de novo protein design marks a shift from prediction to creation — from using computational tools to understand existing proteins to engineering entirely new ones from scratch.
Traditional protein prediction models, like AlphaFold2, have demonstrated extraordinary accuracy in determining the three-dimensional structure of naturally occurring proteins based on their amino acid sequences. This has revolutionized fields like structural biology and enabled faster drug target validation. However, de novo protein design represents a more radical frontier. Instead of asking, "What does this natural sequence fold into?" researchers using AI now ask, "What sequence do I need to build a protein with entirely new properties?"
De novo protein design means creating proteins with no precedent in nature: novel shapes, functions, and binding capabilities unconstrained by evolutionary history. While nature’s proteins evolved under pressures of survival and reproduction, therapeutic proteins have different goals: enhanced specificity, higher stability, reduced immunogenicity, and optimal pharmacokinetics. AI systems excel at exploring this vast design space, where possible proteins number in the millions or even billions — far beyond what human designers or laboratory screening could manage alone.
The ability to generate synthetic proteins purpose-built for drug development promises not only speed but performance advantages that natural proteins may not offer. AI can optimize new proteins for improved binding to disease targets, resistance to degradation in the body, and better compatibility with delivery systems, like nanoparticles or gene therapy vectors. In doing so, AI is enabling a new generation of biologic therapeutics that are not limited by the imperfections or compromises of natural evolution but instead built for the demands of modern medicine.
Understanding the Science – How AI Designs Proteins from Scratch
Learning the Protein Code
The design of de novo proteins by AI is grounded in the ability of AI tools to learn and generalize from massive biological datasets. Just as language models like ChatGPT learn to predict words in a sentence, protein design models are trained to predict and generate sequences of amino acids that fold into desired three-dimensional structures. This is possible because of the exponential growth of protein data, including structure-function relationships cataloged in databases like the Protein Data Bank and sequence data generated through high-throughput genomics and proteomics initiatives.
Early breakthrough tools, such as AlphaFold2, demonstrated the power of deep learning to predict protein structures from natural sequences with remarkable accuracy.1 Models like AlphaFold use evolutionary sequence data, multiple sequence alignments, and structural constraints to infer the most likely fold of a given protein — a task that had challenged scientists for decades.
Building on this predictive foundation, newer AI systems leverage more advanced machine learning strategies to move beyond structure prediction into protein generation. These include protein-specific language models, trained on millions of sequences to learn the "grammar" of proteins; diffusion models, which iteratively refine a noisy starting structure into a stable, functional fold; and reinforcement learning, which allows systems to optimize protein designs based on performance goals set by researchers.2–4
These tools allow AI not just to understand how natural proteins work but to design synthetic proteins that have never existed, with properties optimized for therapeutic applications.
Key Technologies and Platforms
Several platforms now define the state of the art in AI-driven protein design, each with specific strengths depending on whether the goal is prediction or creation.
AlphaFold2 and its successor AlphaFold3 remain essential tools for predicting the structures of natural and engineered protein sequences. These models have provided structural insights into the human proteome and beyond, accelerating research into disease mechanisms and drug targets.1,5
In the realm of generative design, tools like RFdiffusion apply diffusion models to generate completely novel proteins, including enzymes, binders, and scaffolds with high stability and target specificity. RFdiffusion enables the creation of monomers, symmetric oligomers, and interface designs for protein-protein interactions with unprecedented precision.6 Similarly, VibeGenintroduces a dual-model framework to design proteins with specific dynamic properties, expanding the potential to engineer proteins with tailored mechanical or allosteric behaviors.3
Emerging platforms, such as Copilot by 310.ai and DeepSeq.AI, represent a new wave of accessible tools that bring advanced protein design capabilities to non-specialists. Copilot allows users to specify protein design goals in natural language prompts, dramatically lowering the barrier to entry for researchers in drug development.7 DeepSeq.AI, meanwhile, integrates protein design into experimental workflows at institutions like the University at Albany, enabling faster iteration from in silico models to laboratory validation.8
What’s New and Powerful About These Tools
These AI tools have introduced capabilities that were previously impossible or highly impractical with traditional protein engineering methods. Most notably, they deliver dramatic speed improvements. Protein design cycles that once took months or years of experimental screening can now be compressed into days or weeks of computational modeling.9,10
Beyond speed, AI tools offer creativity unconstrained by natural evolution. Whereas nature’s proteins are the product of billions of years of adaptation to environmental pressures, AI can explore entirely new regions of protein space — creating folds, interfaces, and functions never seen before.11,12 This allows for the engineering of proteins that are optimized for specific therapeutic challenges, such as enhanced binding to elusive drug targets or resistance to enzymatic degradation in the body.
Finally, these tools enable unprecedented customization. AI-generated proteins can be designed with built-in features like reduced immunogenicity, improved solubility, or compatibility with specific drug delivery platforms.13 Companies like Generate Biomedicines are leveraging these capabilities to create next-generation therapeutics that are not only more effective but also more manufacturable and scalable than their natural counterparts.
This combination of speed, creativity, and customization positions AI-driven de novo protein design as one of the most transformative developments in modern drug discovery — unlocking new therapeutic modalities that were once beyond reach.
Early Milestones and Missed Shots – A Short History of Progress
First Breakthroughs
The modern era of AI-driven protein science began not with design but with prediction. AlphaFold2, developed by DeepMind, stunned the scientific community in 2020 by accurately predicting the three-dimensional structures of proteins across a wide range of organisms — a grand challenge in biology that had resisted solution for 50 years.5 By demonstrating that deep learning models could infer protein folding from amino acid sequences with near-experimental accuracy, AlphaFold2 became a foundational technology not only for basic research but also for drug development.9
While AlphaFold2 unlocked a new understanding of natural proteins, it also laid the groundwork for a more ambitious goal: de novo protein design. At the forefront of this effort was David Baker's lab at the University of Washington, which applied AI techniques to design enzymes, miniproteins, and synthetic binders with specific functions not found in nature. Baker's innovations combined structural prediction with generative design, producing synthetic proteins with therapeutic potential that could not have been discovered through evolutionary search alone.14,15
These advances found real-world relevance during the COVID-19 pandemic, when AI models were rapidly adapted to design antiviral proteins targeting SARS-CoV-2. DeepMind and other groups developed synthetic binders and inhibitors based on computationally generated structures, illustrating the speed at which AI tools could respond to emerging health threats.1,16 While not all these designs reached the clinic, they proved that de novo protein generation could move from concept to candidate in unprecedented timeframes.
Commercial Applications Begin
The early scientific successes of AlphaFold2 and de novo design platforms quickly catalyzed the formation of new companies seeking to bring AI-generated proteins to the clinic. Generate Biomedicines, founded in 2018, emerged as one of the first biotech companies explicitly built around AI-driven protein design, using generative modeling to create novel therapeutics across immunology, oncology, and infectious disease.17
Similarly, somorphic Labs was launched by Alphabet as a drug discovery company leveraging the advances of AlphaFold to predict not only protein structures but also protein–ligand interactions — a critical step toward designing small molecules and biologics with higher success rates.18
These efforts are already advancing drug candidates toward clinical trials. AI-designed proteins have now entered preclinical and early clinical testing, targeting diseases where traditional biologics struggled to achieve adequate specificity, stability, or manufacturability.19,20 While most of these programs remain in their early stages, the pipeline of AI-generated therapeutics is expanding rapidly.
Misses and Missteps
Like any new technology, AI-driven protein design has experienced its share of limitations and failures. One early challenge was the problem of false positives — AI models confidently predicting that a designed protein would fold or function correctly, only for experimental validation to reveal instability or poor performance in biological systems.21 The models' ability to generate plausible structures often exceeded their capacity to predict real-world behavior, especially in the complex environment of living cells.
A related issue is the limitation of in silico validation. While computational models can evaluate many properties of designed proteins, there remains a critical gap between design and experimental reality. Factors such as solubility, posttranslational modifications, and immune responses are difficult to model accurately, necessitating extensive laboratory testing even for promising designs.11,22
One illustrative example of these limitations is the case of AI-assisted antivenom design. While researchers have used AI models to predict toxin structures and design neutralizing proteins, these approaches struggled with the polyclonal complexity of real-world snake venoms, which contain diverse and rapidly evolving toxin repertoires.23 The need to account for broad-spectrum efficacy against multiple venom components highlighted the ongoing challenges of applying AI design tools to highly variable or complex biological systems.
These early missteps serve as a reminder that while AI has introduced powerful new capabilities to protein design, success depends not only on computational innovation but also on rigorous experimental validation and a deep understanding of biological context. As the field matures, the integration of AI with wet-lab workflows will become essential to realizing its full potential.
What AI Is Making Possible That Nature (and Humans) Couldn’t
Building Proteins with No Natural Analogues
Perhaps the most profound impact of AI-driven protein design is its ability to generate entirely new classes of proteins — structures, functions, and mechanisms that evolution never produced. Natural proteins are the product of selective pressures over billions of years, optimized for survival and reproduction rather than the therapeutic needs of human medicine. AI liberates protein design from these evolutionary constraints, enabling the creation of synthetic proteins with novel folds, interfaces, and binding sites specifically engineered for clinical utility.2,24
Unlike traditional biologics, which are often derived from antibodies, hormones, or enzymes found in nature, AI-designed proteins can be built from the ground up to optimize critical pharmaceutical properties. These include improved pharmacokinetics, allowing proteins to persist longer in circulation; reduced immunogenicity, minimizing the risk of triggering unwanted immune responses; and enhanced compatibility with delivery technologies such as nanoparticles, lipid carriers, or viral vectors.13,25 This design flexibility opens the door to next-generation therapeutics that achieve superior efficacy, safety, and manufacturability compared with naturally sourced proteins.
Moreover, because these de novo proteins have no natural precedent, they can be customized with modular domains or functional motifs tailored to specific disease targets — allowing for multifunctional or conditionally activated therapies that go beyond the capabilities of conventional drug modalities.
Enhancing Efficiency and Precision
Another transformative advantage of AI-driven design lies in the compression of the traditional drug discovery timeline. Historically, identifying and optimizing a therapeutic protein required years of experimental screening, mutagenesis, and iterative refinement. AI accelerates this process dramatically by enabling computational exploration of vast sequence and structure spaces before a single experiment is conducted.
This design–test–refine loop can now occur in days or weeks, allowing researchers to generate candidate proteins, evaluate them in silico for desired properties, and then rapidly move the most promising designs into laboratory testing.26 This acceleration reduces both the cost and risk of drug development, allowing companies to explore multiple design strategies in parallel.
AI also introduces new levels of precision through the use of zero-shot and few-shot learning approaches, where models generate highly functional protein designs with minimal training data for a given target.3,4 In practice, this means that AI systems can extrapolate from their general knowledge of protein structure and function to produce novel candidates for rare or emerging disease targets — even when little experimental data exists.
These capabilities are particularly valuable in scenarios, such as pandemic preparedness or rare disease therapeutics, where time and available biological data are limited.
Addressing Previously Intractable Targets
AI’s ability to explore the full landscape of protein space also allows it to tackle drug targets that have historically been considered intractable. Among the most challenging are intrinsically disordered proteins (IDPs), which lack stable structures and play critical roles in diseases like cancer and neurodegeneration. Traditional structure-based drug design has struggled to address IDPs because of their dynamic and flexible nature, but AI models can generate synthetic binders or stabilizing proteins specifically tailored to interact with these elusive targets.10,11
Similarly, AI design tools are expanding the possibilities for modulating protein–protein interactions — a therapeutic frontier with vast potential but notorious difficulty due to the large, flat, and often featureless interfaces involved. By generating synthetic proteins with precisely engineered binding surfaces, AI enables the creation of inhibitors or scaffolds capable of disrupting or stabilizing these interactions with high specificity.
Membrane proteins, including G protein–coupled receptors (GPCRs) and ion channels, represent another class of challenging targets where AI is beginning to make inroads. These proteins are difficult to crystallize or model experimentally, but AI-driven design can produce synthetic components that facilitate their study, stabilization, or therapeutic modulation.
By extending the reach of drug discovery into these difficult biological spaces, AI is not only accelerating the pace of development but expanding its scope — enabling therapeutic strategies that were once unimaginable.
Ongoing Developments to Watch
Major Research Institutions and Collaborations
As AI-driven protein design evolves from conceptual innovation to clinical implementation, a growing number of leading institutions and biotech partnerships are shaping the future of the field. Among them is ETH Zurich, where researchers have advanced AI-generated protein therapeutics into preclinical and clinical development. These efforts include engineered binding proteins and biologics designed entirely in silico, optimized for therapeutic function and manufacturability before entering the laboratory — a notable milestone in translational application of de novo design.27
The Wyss Institute at Harvard University is another epicenter of innovation, serving as both a research hub and incubator for AI-first biotech companies. By integrating AI platforms with experimental infrastructure, Wyss researchers and their partners are accelerating the entire design-build-test-learn cycle, fueling spinouts that rely on AI as a foundational component of their drug development strategy.20 This fusion of academic research, automation, and startup energy reflects a broader trend: AI is not only transforming how biologics are made, but also who is positioned to make them.
Open Source versus Proprietary Technologies
As with many AI breakthroughs, the field of protein design is now navigating tensions between open science and proprietary control. The release of AlphaFold2 as an open-source model in 2021 democratized access to protein structure prediction tools, sparking a wave of innovation across academia and industry. However, its successor, AlphaFold3, has not yet been made publicly available, prompting concern among researchers that key advances may become siloed within corporate platforms.28
In contrast, RFdiffusion has been released as open-source software, allowing other scientists to build upon its algorithms and apply them to new challenges in protein engineering.6 This divergence illustrates an ongoing philosophical divide: should powerful AI tools for biology be treated as public goods, or as commercial assets?
Beyond questions of access, the debate also encompasses reproducibility and scientific integrity. Proprietary models may lack transparency in how they are trained or validated, limiting external scrutiny and slowing collective progress. Open platforms, while more widely accessible, can raise concerns about misuse, quality control, or misinterpretation of results — particularly by those without deep domain expertise.14,29 Resolving this tension will shape how rapidly and equitably AI-designed proteins become part of the global therapeutic arsenal.
Regulatory and Trust Challenges
As AI moves from the lab into clinical settings, regulators and developers face new questions around accountability, safety, and interpretability. Many AI models function as “black boxes,” generating designs that appear to work in silico but offer little transparency into how or why specific sequences were chosen. This lack of explainability creates barriers to regulatory approval, where detailed understanding of a therapeutic’s mechanism of action and risk profile remains paramount.22
In response, there is growing advocacy for explainable AI — systems that not only generate effective designs but also provide interpretable rationales for their decisions. Achieving this will require new standards for model transparency and validation, including frameworks for comparing AI-designed proteins to natural analogues and benchmarking predicted performance against experimental outcomes.
Other regulatory hurdles include intellectual property (IP) protection for AI-generated sequences. Determining who owns an AI-designed protein — the model creator, the user, or the training data contributors — remains a murky area of law. There is also the challenge of establishing safety testing protocols for proteins with no evolutionary precedent, particularly when traditional animal models may not adequately predict human responses.14,17,21
These unresolved issues underscore a critical reality: the pace of AI innovation has outstripped the frameworks traditionally used to govern drug development. Addressing these gaps will require collaboration between developers, regulators, and bioethicists to ensure that the benefits of AI-driven design are realized responsibly and equitably.
The Road Ahead – What’s Next for AI-Driven Protein Therapeutics?
More than Just Drug Discovery
While much of the attention around AI-driven protein design has focused on therapeutics, the potential applications extend far beyond traditional drug discovery. AI models capable of designing de novo proteins are now being applied to a growing set of challenges in biotechnology, including vaccine development, diagnostics, biosensor creation, and industrial enzyme engineering.30,31
In the field of vaccines, AI-generated immunogens can be designed to elicit stronger, broader, or more durable immune responses than naturally derived proteins, a particularly valuable tool in the fight against rapidly mutating pathogens like influenza or SARS-CoV-2. Diagnostic tools can be enhanced with AI-designed biosensors that recognize disease biomarkers with ultra-high specificity and sensitivity, expanding the reach of personalized medicine.
Perhaps most exciting is the prospect of designing proteins that go beyond static functions. AI models can now engineer switchable proteins that change conformation or activity in response to environmental cues, enabling condition-responsive therapeutics. Multispecific proteins — capable of binding multiple targets simultaneously — offer new ways to modulate complex disease networks, while synthetic scaffolds can be tuned to regulate biological processes in highly precise ways.19,32 These capabilities signal a future where protein therapeutics are not only more effective but also more dynamic and adaptable than ever before.
From Generative Models to Closed-Loop Wet Lab Integration
As AI tools grow more sophisticated, a key development to watch is the increasing integration of computational models with automated laboratory infrastructure. This "closed-loop" approach pairs generative protein design models with robotic platforms capable of synthesizing, expressing, and testing thousands of candidate proteins in parallel — dramatically accelerating the design–build–test cycle that underpins drug development.11,26
Real-time iterative feedback between computational predictions and experimental results enables continuous refinement of AI models. This closed-loop integration reduces the gap between in silico design and biological reality, ensuring that the proteins generated by AI are not only theoretically optimal but also experimentally validated and manufacturable at scale.5,20
Several institutions and companies are already building such platforms, combining high-throughput DNA synthesis, cell-free protein expression, automated assay systems, and machine learning optimization. This infrastructure is essential not only for rapid therapeutic development but also for enabling more ambitious protein engineering goals, such as the creation of entirely synthetic metabolic pathways or programmable cellular behaviors.
Visions of the Future of Programmable Biology
Looking further ahead, the convergence of AI, synthetic biology, and advanced manufacturing is poised to unlock a new era of "programmable biology." In this vision, proteins — and by extension, cells and organisms — can be designed, built, and controlled with the same precision as digital code.1
One frontier area is the development of personalized de novo protein therapeutics. AI models could eventually generate bespoke proteins tailored to an individual’s unique genetic makeup, immune profile, or disease state — an approach with transformative potential for oncology, rare diseases, and immune disorders.19
Another emerging trend is the intersection of protein design with cell therapy and gene therapy. Synthetic proteins may be used to control cell fate, regulate gene expression, or enhance the function of engineered immune cells. The ability to design proteins that interact predictably with synthetic genetic circuits or programmable cell systems represents a major step toward the fusion of therapeutic biologics with synthetic biology.4
Ultimately, the long-term vision is not just faster drug discovery but a redefinition of biology itself — a shift from observing and modifying natural systems to engineering entirely new biological solutions from first principles. AI-driven protein design is one of the key technologies making that future possible.
Conclusion – A Paradigm Shift in Drug Design
The emergence of AI in de novo protein design marks more than a technological breakthrough — it represents a fundamental reimagining of what drug discovery can be. No longer confined to modifying nature’s blueprints, scientists can now craft new proteins from first principles, tailored to meet the needs of human health rather than the constraints of evolution. AI-driven platforms are accelerating discovery timelines, enhancing precision, and expanding therapeutic possibilities far beyond what traditional biologics and small molecules could offer.
From AlphaFold’s structural insights to generative models like RFdiffusion and VibeGen, the field has moved rapidly from predicting protein shapes to designing entire classes of proteins that never existed before. This transition has not only brought new therapies within reach but has also reshaped the ecosystem of biotech innovation, democratizing access through open-source tools while raising critical questions about regulation, ownership, and trust.
Yet, for all its power, AI is not a substitute for biology — it is a partner. The most significant challenge ahead is not generating sequences or folds but understanding and trusting the proteins we create. Experimental validation, ethical oversight, and interpretability must evolve alongside technical capability to ensure that these powerful tools translate into safe, effective, and equitable treatments.
As AI continues to advance, the promise of programmable biology becomes less speculative and more actionable. We are entering an era where the boundaries between computation and life science are dissolving — and in their place, a new paradigm of rational, rapid, and radically creative drug design is taking shape.
References
1. “AlphaProteo generates novel proteins for biology and health research.” Google DeepMind. 5 Sep. 2024.
2. Winnifrith, Adam, Carlos Outeiral, Brian Hie. “Generative artificial intelligence for de novo protein design.” arXiv. 15 Oct. 2023.
3. Ni, Bo and Markus J. Buehler. “Agentic End-to-End De Novo Protein Design for Tailored Dynamics Using a Language Diffusion Model.” arXiv. 14 Feb. 2025.
4. Tang, Xiangru, et al. “A survey of generative AI for de novo drug design: new frontiers in molecule and protein generation.” Briefings in Bioinformatics. 25: bbae338 (2024).
5. King, Anthony. “Four ways to power-up AI for drug discovery.” Nature. 27 Feb. 2025.
6. “RFdiffusion now free and open source.” Institute for Protein Design. 30 Mar. 2023.
7. “Copilot.” GenAI + 31O. Accessed 15 Apr. 2025.
8. Rulson, Larry. “UAlbany professor works with AI firm on drug discovery research.” Times Union. 29 Dec. 2024.
9. Trafton, Anne. “Analyzing the potential of AlphaFold in drug discovery.” MIT News. 6 Sep. 2022.
10. Branca, Malorye. “The Art of AI in Drug Discovery.” Inside Precision Medicine. 12 Feb. 2024.
11. Fu, Chen and Qiuchen Chen. “The future of pharmaceuticals: Artificial intelligence in drug discovery and development.” Journal of Pharmaceutical Analysis. 26 Feb. 2025.
12. Qiu, Xinru, et al. “Advances in AI for Protein Structure Prediction: Implications for Cancer Drug Discovery and Development.” Biomolecules. 14: 339 (2024).
13. Hashemi, Samaneh, et al. “Therapeutic peptide development revolutionized: Harnessing the power of artificial intelligence for drug discovery.” Heliyon. 10: e40265 (2024).
14. “AlphaFold 3, AI Tools in Drug Discovery, and Patentability.” Pharmaceutical Law Group. 5 Mar. 2025.
15. Chan, Kelvin, Christina Larson, and Manuel Valdes. “Nobel Prize in chemistry honors 3 scientists who used AI to design proteins — life’s building blocks.” AP. 9 Oct. 2024.
16. Eisenstein, Michael. “AI-enhanced protein design makes proteins that have never existed.” Nature Biotechnology. 31: 303–305 (2023).
17. Shah-Neville, Willow. “12 AI drug discovery companies you should know about.” Labiotech. 27 Mar. 2025.
18. “Isomorphic Labs.” Accessed 15 Apr. 2025.
19. Sun, Duxin and Christian Macedonia. “Will AI revolutionize drug development? Researchers explain why it depends on how it’s used.” JHEOR. 7 Jan. 2025.
20. Alucozai, Milad, Will Fondrie, and Megan Sperry. “From Data to Drugs: The Role of Artificial Intelligence in Drug Discovery.” Wyss Institute. 9 Jan. 2025.
21. Paul, Debleena, et al. “Artificial intelligence in drug discovery and development.” Drug Discovery Today. 25: 80–93 (2020).
22. Lu, Jiankyn. “AI could accelerate drug discovery. But only if we can trust it.” The Rockefeller University. 16 May 2024.
23. “AI Is Coming Up With Brand New Molecules, Fueling Drug Discovery.” Science Friday. 24 Jan. 2025.
24. “AI designs new drugs based on protein structures.” ScienceDaily. 24 Apr. 2024.
25. Ocana, Alberto, et al. “Integrating artificial intelligence in drug discovery and early drug development: a transformative approach.” Biomark. Res. 13:45 (2025).
26. Buntz, Brian. “2024: The year AI drug discovery and protein structure prediction took center stage—2025 set to amplify growth.” Drug Discovery and Development. 25 Nov. 2024.
27. Bergamin, Fabio. “AI designs new drugs based on protein structures.” ETH Zurich. 24 Apr.2024.
28. Lane, Alasdair. “The AI-Powered Future of Drug Discovery.” The Atlantic. Accessed 15 Apr. 2025.
29. “AI Method for Generating Proteins Will Speed Drug Development.” Lab Manager 30 Mar. 2021.
30. Schubert, Charlotte. “How AI is changing the way scientists engineer drugs, biosensors, enzymes and more.” GeekWire. 10 Jul. 2023.
31. Özçelik, R, et al. “Structure-Based Drug Discovery with Deep Learning.” ChemBioChem. 24: e202200776 (2023).
32. Chung, Jason, et al. “Artificial Intelligence: A New Tool for Structure-Based G Protein-Coupled Receptor Drug Discovery.” Biomolecules. 15: 423 (2025).