A Blueprint for Breakthroughs
Decentralised science (DeSci) marks a new era for discovery by breaking down institutional and hierarchical barriers. When coupled with AI, R&D will be superpowered.
That’s right, DeSci and AI are a match made in innovation heaven. Read on to discover precisely why these two technologies create such powerful synergies.
Introduction
DeSci is ripping up the rulebook for bio. By throwing open the doors to scientific knowledge and harnessing the power of open-source, it’s unleashing a tidal wave of credible new approaches and collaborations. The result? A total disruption of the typical go-to-market model for the drug development and biotech industry.
The problem: Pharma’s ROI on R&D is in freefall. Skyrocketing costs, snail-pace development, and sky-high abandonment rates, leaving many potential breakthroughs unrealised.
The solution: By unleashing protocol tokens and inventive incentive schemes, DeSci’s sparking collaboration like never before. Goodbye competition, hello cooperation! Enhanced public welfare and investor returns. Barrier-free and global, it reveals untapped scientific avenues and talent.
Following our previous exploration of DeSci’s impact on the pharmaceutical and academic worlds, we now examine AI’s role in elevating DeSci’s potential. The integration of DeSci and AI is not just groundbreaking but essential.
In this analysis, we’ll peel back the layers of the key DeSci frameworks, protocols, and platforms driving this evolution, and delineate AI’s vital contribution.
By the end, I hope the diagram below is clear to my readers! A visual roadmap to encapsulate the core concepts and mechanisms that we’ll progressively unpack throughout the following sections — a compass to guide us through the exciting terrain ahead.
AI is not just an enhancer to DeSci drug discovery but also a solution to reducing the risk of failure in human trials.
AI is Transforming Biotech
AI’s role in modern biology is becoming clear, syncing structural bioinformatics with the genomic revolution. Dr. Shelby’s two articles eloquently chronicle AI’s game-changing impact on drug development.
From medicine design to discovery,, screening to clinical development, AI is leaving its mark at every stage. It’s deepening our understanding of biology, revolutionising genetic analysis, optimising molecules, and paving the way for personalised treatments. Even medical literature review is getting an AI upgrade.
The industry isn’t just taking notice; it’s embracing AI with open arms. Pharmaceutical giants and nimble biotechs alike are weaving AI into their pipelines at an astonishing pace. We’re witnessing AI-designed drugs not just in early stages, but in advanced human trials!
Let that sink in for a moment. Molecules dreamed up by artificial minds are being tested in real patients, edging closer to market with each passing day.
So, what are the types of artificial intelligence that play the prominent roles within drug discovery:
Variational Autoencoders (VAEs)
VAEs simplify high-dimensional data — converting the complex into a more understandable form. They not only generate new molecular structures but also refine drug properties for better efficacy and safety — a skilled chemist with an infinite imagination. VAEs manage complex datasets at lower computational costs versus other AI models. This turbo-charged performance is a game-changer for tackling urgent health challenges.
Diffusion Models
To parallel genomic sequencing advancements, diffusion models bypass slow traditional methods of bioinformatics (reducing the reliance on wet labs). They leverage vast databases for faster, more accurate protein predictions.
But the benefits don’t stop at the molecular level. A better understanding of biology means human studies will be more efficacious and cheaper. They promise more efficient, cost-effective clinical trials and personalised treatment designs by analysing individual genetic responses to drugs. It’s a world where treatments are tailored to your DNA, not just your symptoms.
Large Language Models (LLMs)
In early drug discovery, generalised LLMs excel in tasks that demand an expansive understanding of biological mechanisms, demonstrating proficiency across various bioinformatics and healthcare problems.
They also hold promise for addressing wider industry challenges. By adeptly handling the integration of diverse information, they “think” in ways the other models can’t — ideal for identifying new research areas, generating hypotheses, and summarising publications.
Generative Adversarial Networks (GANs)
Imaginative powerhouses the produce high-fidelity synthetic images, such as MRIs and CT scans, GANs enable AI-model training without extensive real-patient datasets, crucial in scenarios where patient data is scarce (e.g. rare diseases). In molecular design, GANs can be conditioned for more precise control compared to other models discussed.
Multi-Agent Systems (MASs)
An accompanying essay, Serendipity on Steroids, described how multiple-LLMs are already autonomously conducting intricate chemical experiments, and powering robotic laboratories for engineering enzymes.
The essay expands upon the role of Blockchain powered MASs, provided by the likes of OLAS and DAIN, in streamlining the discovery of patentable innovations. It delves into their ability to make genuine scientific breakthroughs at unprecedented speed.
These AI-driven breakthroughs are not just incremental improvements; they’re giant leaps forward in our understanding of disease and our ability to treat it.
DeSci Primitives and AI Integrations
The advent of blockchain technology has introduced novel and exotic instruments that will completely change the manner in which research is conducted. Reflecting on the twilight of my tenure in drug development, an realisation dawned upon me. There lurked a mysterious force within the industry, more potent than any other… but what was this force?
It wasn’t the pharmaceutical giants.
It wasn’t the regulatory authorities either.
The true power lay with Patient Advocacy Groups. These organisations, forged through the trials of serious health conditions, distinguished themselves with their ingenuity and unwavering will. They boasted robust funding, exceptional organisation, and a drive to eclipsed all others.
Their understanding of their ailments often outstripped that of the clinicians treating them, showcasing a remarkable example of collective intelligence, a.k.a. a hive mind. Big Pharma often found itself acquiescing to their influence, underscoring their capacity to challenge and even reverse regulatory verdicts.
Unlike many superficial DAOs of the crypto-world, which are DAOs in name only, these groups were spiritually DAOs by their very nature. They were the proto-bioDAOs.
Decentralised Autonomous Organizations (DAOs):
At the vanguard of biomedicine’s transformation, BioDAOs like Vita, Valley, Athena, Hair, Cerebrum and Cryo are redefining research from the ground up. Tokenised models intricately weave individual ambitions with the collective goal of fostering public goods, drawing inspiration from Nimbus’s pioneering decentralised drug development and pharma licensing framework.
These entities thrive as meritocracies, where the value of contributions transcends monetary input. The introduction of the BIO token epitomises this principle, enhancing DeSci’s ecosystem by enriching participation, governance, and liquidity. Anchored by the collaborative platforms of DeSci Labs and LabDAO, they champion a culture of interdisciplinary collaboration, transparency and equity, with discoveries meticulously documented on-chain for the benefit of all.
They may get criticised for being open to governance attacks but new token standards are emerging to mitigate such issues. Paired with reputation systems (more on that later) they will be supercharged.
AI’s burgeoning role in biotech, underscored by 2023’s seminal advancements in protein science (ProGen2, AlphaFold, ESMFold and RoseTTAFold), is setting the stage for AI to seamlessly integrate into DAO frameworks. Yet, the often proprietary nature of these insights and the global scarcity of computational resources limit access to these revolutionary findings. The public domain’s lack of training data for advanced models further exacerbates this issue.
Projects such as Inference Labs decentralise the models, while others like Prime Intellect aggregate the compute markets. This enables DAOs to cultivate rich, diverse datasets that truly represent our heterogenous global population. “Minorities” in the genebanks of the west are not minorities around the world.
DAOs can use protocols like Data Lake to democratise access to anonymised patient data, ensuring that resultant innovations are shared equitably across humanity. Before us stand:
- Therapeutic area specific fine-tuned LLMs, owned and monetised by DAOs.
- DAO personalised treatments using Diffusion Models trained on members’ data, as well as synthetic GAN data.
Research shows that such fine tuned models can be trained with less resources and still outcompete the leading incumbent models. BioDAOs will fight back against the consolidation of power around those who are data and compute rich. They will level the playing field.
Reputation Systems and Decentralised Identity (DIDs):
These systems are crucial for fostering trust and collaboration between humans and autonomous agents alike. When it comes to drug discovery, the consequences of acting on false information can be severe. Such systems log the credibility and reliability of DAO participants with identity proofs that can be minted privately on the Holonym network for key and data custody.
Mechanisms for recognising and rewarding high-quality contributions push participants to strive for excellence. They drive the overall quality of research and innovation higher.
A guiding principle from my former Chairman, “Drugs cannot be designed by a committee,” echoes through the corridors of discussions on BioDAOs when I talk to outsiders. Nevertheless, the deployment of token-gated reputation systems, such as those being developed by ResearchHub, elegantly counters the scepticism towards their egalitarian structures.
Through these systems, pivotal decisions are reserved for the wisdom of distinguished experts who traverse the landscape of BioDAOs, imparting invaluable insights and garnering compensation that mirrors their verifiable and on-chain endorsed reputations. Other DAO members delegate their votes to those who know what they are doing.
For autonomous agents, reputation frameworks offer a method to gauge their contributions and interactions, instilling confidence in wary human scientists through rigorous vetting. The sanctity of data, especially for AI model training, cannot be overstated. Ceramic “streams” provide a flexible means to embody all manner of data a DAO AI model could necessitate, from research data and patient data to documents and schemas.
Decentralised Knowledge Graphs (DKGs):
Graph machine learning holds transformative potential for drug development through its ability to analyse large scale biological networks. By understanding the intricate relationships and dynamics within these systems, researchers can identify novel targets for therapies.
While blockchains are excellent for building trust their capabilities in handling dynamic databases are limited. DKGs bridge this gap. OriginTrail leads this charge, crafting verifiable DAO-based knowledge graphs for machine learning and associated knowledge markets to facilitate discovery across critical fields.
BenevolentAI is mining the scientific literature to create knowledge graphs for drug discovery — all within a proprietary environment. In contrast, DKGs address the inherent limitations of traditional data silos through transparency, ownership and data composability. They are DeSci’s ‘Google,’ offering agile, query-friendly environments for autonomous agents to roam.
AI algorithms harness the intricate datasets from DKGs, extracting insights with unprecedented velocity and depth, surpassing the capabilities of conventional biotech approaches. These graphs guide autonomous agents in discerning trends and associations that may bypass human observation. Pure hypothesis generators.
This ability of AI to continuously integrate new findings into thinking ensures that the latest research continuously furnishes the broader scientific understanding. DKGs are a cornerstone for constructing Tech Trees and Outcome Graphs, essential tools for understanding technical capabilities, identifying challenges, and pinpointing impactful areas.
Forward-thinking organisations such as the Foresight Institute employ these analytical frameworks to explore and advance the frontier realms of longevity, space, and nanotechnology, marking a trail for future explorations.
Hypercerts:
Hypercerts revolutionise incentives for contributions, shifting from a culture of knowledge hoarding to one of sharing and collaboration. By offering retroactive rewards for both groundbreaking and foundational research, they promote all-encompassing transparency in research endeavours.
Through Coordination.network these groundbreaking primitives are woven into therapeutically focused tech trees, crafting an impactful “push mechanism” that synergises with the established “pull mechanisms” of patents and economic incentives. Importantly, they encourage the dissemination of both positive and negative results — the latter is generally never published — reducing repetition in experiments and with it environmental impact.
Within these frameworks, LLMs assume a pivotal role, identifying and assessing nascent contributions that could accelerate scientific breakthroughs. Uncovering underappreciated insights and incentivising people to execute on them. Together, AI and Hypercerts unlock a synergistic cycle: AI’s analytical capabilities amplify the effectiveness of Hypercerts, which in turn create an open and data-rich environment in which AI can thrive.
This model ensures that contributors of foundational research, the bedrock of future medicinal advancements, are duly rewarded.
Quadratic Funding:
BioDAOs can leverage other interesting crypto-enabled mechanisms such as Quadratic Funding by valuing the number of supporters over the size of donations. Gitcoin consolidates all donations, large and small, into a matching pool and ensures projects with more numerical support receive more, championing the voice of “the many” over “the affluent few.”
Together with hypercerts these primitives construct a dense network of community-driven insights that feed into DKGs and guide resources to where they’re most valued.
Prediction markets:
Crowdsourcing’s potential stretches beyond mere funding and out into the realm of forecasting. Prediction markets permit speculation on an asset’s efficacy or clinical outcomes, well before traditional timelines could foresee.
By weaving IP-NFT and asset prediction markets into BioDAOs we significantly enhance the agility and accuracy of decision-making in drug development, tapping into the collective ‘wisdom of the crowd’. Not only do these markets allow AI to streamline resource allocation and risk assessment, they further attract a spectrum of speculators eager to capitalise on shorter term investment cycles.
Envision a world where AI multi-agent systems enriched by vast interconnected DKGs actively seed and navigate these markets, catapulting drug discovery forward. This scenario is not just fantasy; it’s our present reality, underscored by Vitalik Buterin’s observations on the recent engagement of Olas agents within prediction markets like Omen.
For a more comprehensive explanation of how autonomous agents will revolutionise drug development, please read serendipity on steroids.
Intellectual Property NFTs (IP-NFTs) and Tokens (IPTs):
Molecule’s pioneering IPTs and NFTs are transforming IP creation, management and exchange. Traditional inefficiencies are eliminated by merging open infrastructure with smart contract-encoded rules and rights. They align early speculative investors, intrinsic value investors, patent maintainers, and other stakeholders around an on-chain IP asset.
Researchers entering into a research agreement contribute their findings to a DAO, securing a stake in the intellectual property that is subsequently filed and represented as an IP-NFT. The DAO, as the financier of the research, holds a portion of the IP-NFT and has the flexibility to auction fractions of this tokenised asset (IPT) to generate funds or incentivise certain activities.
These new tools stimulate community engagement in science, embedding incentives for usage rights and revenue sharing, and connect IP to DeFi, expanding financial opportunities for inventors (patents and data can now be used as financial collateral).
Importantly, scientists can now own their innovation directly. They can monetise it at various stages to bootstrap follow-on work and make their patent estate more robust. Reputation systems can be used to avoid a “Commons of Knowledge” dilemma, allowing co-development shrouded in the open.
In simple terms, IPTs open up secondary markets. Investment possibilities that were previously the sole domain of biotech VCs and PE firms reach a wider range of investors. They permit continuous fundraising to match the long lifetime and evolving value of IP assets. Rather than a one-time capital raise, investment can be drawn in over time as value becomes more certain.
Utilising diffusion models trained on BioDAO patient data, agents can identify effective treatments for members. They can uncover new applications for existing medications, or breath life into underdeveloped assets from struggling biotechs. This mirrors the strategic repurposing of assets by entities such as Roivant, BridgeBio and Cambrian Biopharma.
In our work at GW Pharma, we focused on developing drugs from natural compounds. Given that naturally occurring entities cannot be patented, we established barriers for competitors through a comprehensive matrix of use, manufacturing, and formulation patents. It’s not that each was individually insurmountable, but when stacked up it became unattractive.
IP-NFTs, with their composable and interoperable characteristics, are ideally suited for constructing such elaborate protective frameworks. They simplify what used to be a convoluted bureaucratic nightmare.
The Blueprint:
We’re at 2000 words now, so here’s a picture to save you another 1000:
Conclusion
The radical advancements in AI capabilities are reshaping an already uncharted DeSci landscape, opening up questions abound what to fund. The clinical phase of drug development traditionally spans 8–10 years, but initiatives like Vitalia — a “Los Alamos” for longevity biotech — promise to accelerate drug development to warp speed (as we saw during Covid), via Special Economic Zones. It is reminiscent of Dubai’s transformation in 1970, but centered on science instead of finance, and will leverage regulatory arbitrage opportunities within DeSci.
With the potential to compress the traditional decade-long journey of drug development into significantly shorter timelines, the role of humans evolves from hands-on execution to strategic decision-making.
This shift underscores the preservation of human agency in an increasingly automated world, granting us the power to shape the future we envision. DeSci offers an open framework to direct AI, serving as aligned architects of a future designed with our collective aspirations in mind.
Thank you to Paul Kohlhass, Tyler Golato, Shawn Dimantha, Vincent Weisser, Shady El Damaty, Laurence Ion, Warren Winter and Martin Karlsson for their review and valuable feedback during the writing of this essay.
ID Theory may hold positions in some of the projects discussed in the above. This article is strictly for informational and educational purposes only. It does not in any way constitute an offer or solicitation or an offer to buy or sell any investment or cryptoassets discussed herein. Always perform your own research and conduct independent due diligence prior to making any investment decisions.
Interested in partnering with ID Theory or building something special? Get in touch through our website or at info@idtheory.io.
References:
- Editorial (2024). Tearing up the traditional biotech playbook. Nature Biotechnology, 42, 1.
- Shelby, D. (2022). The ‘recalibration phase’ of AI in drug discoverySubstack. URL: https://shelbyann.substack.com/p/bio-foundation-models-definitions
- Shelby, D., Boniolo, F. (2023). Bio foundation models — definitions, nuance, and tactics. Substack. URL: https://shelbyann.substack.com/p/bio-foundation-models-definitions
- Madani, A., Krause, B., Greene, E.R., et al. (2023). Large language models generate functional protein sequences across diverse families. Nature Biotechnology, 41, 1099–1106.
- Arnold, C. (2023). Inside the nascent industry of AI-designed drugs. Nature Medicine, 29, 1292–1295.
- Microsoft Research. (2023). The Impact of Large Language Models on Scientific Discovery: a Preliminary Study using GPT-4. arXiv.
- Sarwal, V., Munteanu, V., Suhodolschi, T., Ciorba, D., Eskin, E., Wang, W., & Mangul, S. (2023). BioLLMBench: A Comprehensive Benchmarking of Large Language Models in Bioinformatics. arXiv.
- He, K., Mao, R., Lin, Q., Ruan, Y., Lan, X., Feng, M., & Cambria, E. (2023). A Survey of Large Language Models for Healthcare: from Data, Technology, and Applications to Accountability and Ethics. arXiv.
- Skandarani, Y., Jodoin, P.-M., & Lalande, A. (2023). GANs for Medical Image Synthesis: An Empirical Study. Journal of Imaging, 9(3), 69.
- Nijkamp, E., Ruffolo, J. A., Weinstein, E. N., Naik, N., & Madani, A. (2023). ProGen2: Exploring the boundaries of protein language models. Cell Systems, 14(11), 968–978.e3
- Jumper, J., Evans, R., Pritzel, A., Green, T., Figurnov, M., Ronneberger, O., Tunyasuvunakool, K., Bates, R., Žídek, A., Potapenko, A., et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature, 596, 583–589.
- Lin, Z., Akin, H., Rao, R., Hie, B., Zhu, Z., Lu, W., Smetanin, N., Verkuil, R., Kabeli, O., … Rives, A. (Year). Evolutionary-scale prediction of atomic-level protein structure with a language model. Science, 379(6637).
- Baek, M., McHugh, R., Anishchenko, I., Jiang, H., Baker, D., DiMaio, F., et al. (2023). Accurate prediction of protein–nucleic acid complexes using RoseTTAFoldNA. Nature Methods, 21, 117–121.
- Liu, T., & Low, B.K.H. (2023). Goat: Fine-tuned LLaMA Outperforms GPT-4 on Arithmetic Tasks. arXiv.
- Surowiecki, J. (2004). The Wisdom of Crowds: Why the Many Are Smarter Than the Few and How Collective Wisdom Shapes Business, Economies, Societies and Nations. Doubleday. ISBN 978–0–385–50386–0.
- Buterin, V. (2024, January 30). The promise and challenges of crypto + AI applications. Vitalik Buterin’s website. URL: https://vitalik.eth.limo/general/2024/01/30/cryptoai.html
