What we’re looking for in biotech and techbio
Apr 25, 2025
Ashley Lin
At Necessary, we’re excited about platform biotech companies. The ones we’ve been most excited by generally have the following attributes:
Robust computational platforms: i.e., screening large protein/chemical libraries at higher throughput than current methods, predicting ligand-receptor binding, using AI/ML with proprietary data to discover and/or design novel targets, small molecules, and/or proteins. We’re not disinterested in gene and cell therapies, but they’re much lower priority.
Scaled wet lab systems that have demonstrated increased efficiency over competitors. A preferred example is using lab automation to enable 100X the assays in 1/10th the time, especially assays to provide feedback to computational hits. Other examples include efficient protein production systems and low-cost chemical synthesis processes that unlock scaled exploration and validation.
Strategic indication selection: we like multi-indication assets and indications with significant unmet need. For indications with significant external therapeutic development, we must be convinced of abilities to surpass others and retain market share.
Diverse partnership intent opportunities: teams that generate enough validated hits can and should pursue early partnership or licensing deals to create near-term revenue that can support the development of independent programs vs. relying solely on investors.
Balanced founding team: a founding team that has the required skills and experience on both the science/drug discovery side as well as the business development/dealmaking side.
Quality informative data: We expect data sets to provide proof of concept, whether it be wet lab data only or, in certain instances, additional computationally generated data.
You’ll notice that our areas of interest aren’t being categorized by indications, targets, or types of therapies. While we have opinions about those, our goal here is to emphasize what appears to be a more universal approach with regard to discovery and development that is becoming the industry standard.
A specific subset of the above that we’re excited about are AI/ML technologies that advance the protein discovery flywheel:
Protein synthesis
Digital twins of the protein production process to optimize conditions and scale up efficiently
More accurate and high-throughput simulations to predict protein folding and help guide the design of proteins that fold more accurately
Computational methods to more accurately predict protein developability
Protein sequencing
In-vivo protein sequencing for real-time monitoring of protein synthesis and degradation
Predicting protein-protein interactions and sequencing small molecules
Mining the dark proteome & ligand discovery for intrinsically disordered proteins (IDPs)
Other areas of interest include:
AI/ML technologies for predicting toxicity studies, in vivo data, and clinical trial probability of success (POS)
Novel enzyme discovery
GPCR discovery and GPCR drugs for non-sensory indications
Selectively turning down the immune system or turning immune cells on/off
Selectively turning proteins/DNA/RNA on/off
Lowering in vivo and in vitro study costs
Alternative expression systems using synthetic biology, alternative hosts, or cell-free synthesis to minimize costs and production times for difficult-to-express proteins
Large-scale, low emissions continuous biomanufacturing at a significant scale (i.e., able to have 50+ liters/reactor)
Our thoughts on AI in drug discovery
Software and hardware are both dramatically altering the drug discovery landscape. Love it or hate it, TechBio is a real thing and heralds a new age for the industry, e.g., using advanced computational approaches such as AI/ML, big data analytics, and robotics to drive drug discovery and development. Our team has been investing here since before the term was coined, including Recursion Pharmaceuticals (NASDAQ: RXRX), a pioneer of this field that epitomizes much of this thesis. We remain big believers in the power of computation to accelerate the drug discovery process with some nuance/caveats:
Computation is not everything
Computation (in silico) has so far proven highly unreliable, surprise surprise. Speed to in vitro and in vivo studies and the speed of those studies compounds any computational advantage when fed back into computation. Wet-lab processes are typically manually-intensive and remain a bottleneck. Scaled wet-lab processes that include computational approaches to assays with high throughput can overcome these challenges and then some by increasing predictive power.
The wet-lab isn’t everything either
We’ve seen several startups using AI to screen large chemical or protein libraries against targets. Regardless of in silico or wet-lab hits, predictive models are wrong more often than right. Much like venture investing! Also similar to venture investing, D&D requires enough shots on goal, not just higher quality shots. Companies using AI in screening to differentiate themselves need to demonstrate that they are meaningfully generating data guided by a biological hypothesis with an end goal in mind rather than just generating data for the sake of it. Don’t tell me how much data you’re generating, tell me how you’re feeding it back into your computation and how that is measurably improving your computation.
Self-iterative AI/ML model
Related to data generation: we are particularly interested in companies with self-informed models, such as a self-informed protein design platform where the model is an autonomous cycle of prediction, testing, feedback, and optimization. Specifically, we want to understand how fast a company can complete a cycle and how accurately its model can learn from mistakes and self-correct.
Maximizing hits to maximize value capture
Companies with systems that can generate quality hits can create optionality. Specifically, companies can choose to stay as a hit-generation company and operate as a service model for other pharma (becoming like a CRO/operating on a SaaS-like model), turn their hits into licensing deals, or develop their own programs. A combination turns optionality into a blended risk profile, with multiple ways to win. Which hits end up where is a more complex question than this blog can discuss, but that’s the conversation we like to have vs. choosing a single path.
Using AI/ML to explore untapped bottlenecks along the drug discovery process
Many startups are already using AIs to generate hits, novel molecules, and predict binding affinities. We are interested in companies leveraging AI for unexplored spaces along the drug discovery process. Here are several bottlenecks in the drug development process that remain relatively untapped by AI technologies:
In vivo modeling: Predicting in vivo study results can save significant money, which is required for animal studies. However, the in vivo modeling space is fragmented and remains a significant bottleneck.
Modeling for tissue studies: Patients usually provide a blood or tissue sample in cancer clinical trials, which is burdensome. Additionally, the number of high-quality samples falls short of meeting the overwhelming demand. What if this wasn’t needed and could be simulated?
Modeling toxicity studies: Clinical trials rely heavily on animal testing and human patients for toxicity studies. AI models can be implemented for predictive toxicity modeling, in silico testing, or organ-specific toxicity prediction to accelerate timelines, save costs, and reduce the suffering of clinical trial participants.
Modeling the probability of success (POS) of a clinical trial and indication selection: Indication selection and POS studies are still done manually and remain fragmented processes involving multiple parties, such as in-house groups from biopharma companies, CROs, third-party consultant firms, and data analytics firms. AI can be leveraged to analyze multi-omics data, mine disease areas for potential indications, stratify patient populations, analyze real-world evidence (RWE), and optimize cost and resource allocation to help companies decide on the best indication to commercialize their discovery successfully.
We probably wouldn’t invest in a platform pursuing one of these specific bottlenecks due to the narrowness of that as a wedge and the fragmented nature of the customer set. We’d be much more excited to see a company with a roadmap to developing this tooling for their own programs based on a novel approach to data generation, capture, and usage.
Companies We’re Excited About
These are some companies that are building the future we see, that we’ve learned from, and/or that we’ve enjoyed meeting over the last several years:
Glyphic Biotechnologies (Series A): AI/ML algorithms with nanopore signals to accurately sequence proteins at scale, bypassing the traditional bottleneck of costly, low-throughput proteomics analysis
64x Bio (Series A): Novel high throughput genome engineering in a design loop with computational tools to generate highly optimized and otherwise unattainable mammalian cell lines for next generation therapeutics development, with an emphasis on cell and gene therapies
Relay Therapeutics (NASDAQ: RLAY): Dynamo platform integrates advanced computational modeling with experimental techniques to simulate protein dynamics, allowing the discovery of optimized drug candidates for previously “undruggable” protein targets
A-Alpha Bio (Series A): Computational platforms (AlphaSeq and AlphaBind) measures millions of protein-protein binding affinities simultaneously to predict binding strength for immunocytokine and molecular glue discovery
Peptone (Series A): Oppenheimer platform integrates biophysical data, AI, and ML to model IDPs, identify optimal binding sites, and design structurally dynamic drugs to unlock the therapeutic potential of IDPs, with treatments for diseases such as cancer, inflammation, and fibrosis.
Superluminal Medicines (Series A): Computational platform with a “predict-design-test” architecture targets GPCRs to identify specific protein conformations associated with disease states and design highly specific compounds for challenging protein targets
Profound Therapeutics (Series A): ProFoundtry platform utilizes advanced computational and ML methods to mine for novel drugs and drug targets hidden in the expanded human proteome, resulting in a catalog of tens of thousands of novel proteins that are used in high-throughput experimental assays to develop novel therapeutics
Frontier Medicines (Series A): Frontier Platform integrates chemoproteomics, AI/ML, and a covalent fragment library to unlock druggable hotspots, including proteins with poorly defined structures, to target over 90% of the human proteome for mainly oncology indications
Generate Biomedicines (Series C): Utilizes generative biology to generate custom, novel protein therapeutics computationally (can move from computer to clinic in 17 months), with an emphasis on immunology indications
Asimov (Series B): Combines cutting-edge mammalian synthetic biology, computer-aided design (CAD), and ML to design, simulate, and optimize genetic systems precisely, with applications such as biologics production, lentivirus production, and RNA/cell/gene therapy payload design