At Transcripta Bio, our mission is to decode the language of biology to engineer medicines for patients. To achieve this, we see an important role for artificial intelligence as a partner in scientific discovery.

Progress in science is not just about accurate prediction; it’s about building strong explanations. This is particularly true in discovery research, where we venture into the unknown and the concept of a “ground-truth” answer often doesn’t exist. For language-based AI to be a trustworthy collaborator, its reasoning must be transparent and evaluable in open-ended domains.

This challenge is different from automation. Recently, a class of agentic workflow systems—often referred to as scientific copilots—has emerged. Platforms like Biomni and ToolUniverse accelerate research tasks by orchestrating agents to perform data analysis, providing operational speed to existing workflows.

These systems are already impacting scientific practice. However, we believe a complementary form of AI—one that can propose explanations for complex phenomena in the same way a human scientist does—is just as critical for the future of science.

Our recent work addresses this challenge through LLM Kernel, a framework for evaluating the explanatory reasoning of language models in open-ended scientific domains. This research has been accepted to the NeurIPS 2025 AI for Science Workshop.

Beyond automation and towards explanation

While scientific copilots are becoming effective for automating typical procedures, they don’t address the core challenge of discovery research: evaluating the quality and validity of scientific explanations themselves.

Our work on LLM Kernel takes a different path. Instead of building a faster lab assistant, we aim to create a verifiable scientific partner. The goal isn’t just to get an answer more quickly, but to understand if an explanation is valid by measuring the AI’s reasoning process.

LLM Kernel

In discovery research, we can’t just check an answer against a textbook. So, how can we build evidence for explanations?

Our framework draws on a long-standing principle of scientific reasoning: comparison. Many discoveries emerge from recognizing patterns of similarity or deviation—seeing when phenomena behave alike or when one diverges from the expected pattern. This idea is also present in many common statistical methods that use notions of similarity to reveal structure in data.

In transcriptomics, for instance, we often compare gene expression profiles using metrics like cosine similarity, expecting that drugs with related mechanisms of action will yield similar transcriptional responses and thus cluster together. Kernel-based methods generalize this notion, enabling systematic comparisons across rich scientific data types.

LLM Kernel extends this comparative framework to language models. Instead of a fixed mathematical metric, it leverages an LLM’s capacity to ingest unstructured data and reason through analogies and contrasts.

Given two samples, the model produces:

1. A qualitative reasoning trace — a human-readable explanation of the biological similarities and differences it identifies.

2. A quantitative similarity score (ranging from -1 to 1) — a single value summarizing its judgment of how similar the samples are.

This framework forces the model to commit to a testable conclusion: a numerical score that follows directly from its scientific reasoning. Because the explanation precedes the score, we can evaluate not just what the model concludes, but why.

In discovery research, absolute labels are rare, but relative information is abundant: we often know when one sample behaves more like another. By framing the task as a set of pairwise comparisons, unlabeled samples can be organized by multiple notions of similarity.

Unlike classification, which constrains interpretation to a fixed set of predefined categories, similarity scoring allows explanations to emerge directly from the data. Free from coarse labels such as pathways or mechanisms, the model can explore patterns without bias toward existing frameworks, enabling novel associations.

When the LLM kernel produces similarity patterns that align with known sample relationships, it provides evidence that the model’s reasoning is scientifically meaningful—making explanation itself measurable.

Share

Key scientific findings

We applied LLM Kernel to a large-scale transcriptomics dataset to distinguish between compounds with various molecular effects.

Superior performance

LLM Kernel consistently outperformed standard numerical methods like Pearson and Spearman correlation in identifying biological relationships. We found that performance scaled with the computational budget of the model—giving the model more “thinking time” resulted in better similarity scores!

Dependence on biological knowledge

Ablation experiments revealed that the model’s success depends on access to biological context, not just statistical patterns. When gene names were masked, performance dropped sharply. Additional ablations confirmed that the LLM Kernel relies on prior knowledge of gene function and pathway structure rather than merely reproducing numerical correlations.

Cross-modal reasoning

A unique feature of LLM Kernel is its ability to compare completely different data types. We introduced a novel challenge: score the similarity between a natural language description of a disease (e.g., Alzheimer’s disease) and a numerical gene expression profile from a drug-treated cell line. The method was able to identify compounds with known relevance to specific diseases by recognizing cellular responses that complemented each disease description—an impossible task for traditional numerical methods (see Figure 3 of the paper).

Open questions

A key limitation is that while our framework links each similarity score to an explanation through a causal chain of tokens, we don’t yet know the precise origin of that score. Is it truly the product of the detailed reasoning we see, or does it emerge from implicit computations?

Our finding that scores improve with more “thinking time” suggests performance is tied to the volume of computation—not necessarily the quality of the reasoning itself. Understanding the relationship between computation and the accuracy of produced content is a central question in modern AI research, and LLM Kernel may provide a new tool to investigate it.

The road ahead

Using AI to accelerate science may be the most impactful way we apply intelligence technology. LLM Kernel represents a step toward building auditable and verifiable AI for scientific discovery.

Through pairwise comparison, we transform unlabeled scientific data into a series of testable relationships—small, data-driven judgments that reveal how well the model understands biology. By coupling qualitative reasoning with a quantitative, benchmarkable score, we can begin to open up the “black box” of AI’s reasoning and build trust in its interpretations.

This framework provides a scalable way to evaluate and improve the quality of AI-driven scientific insight, paving the way for systems that can act as true partners in the quest to understand and engineer biology.

For a deeper dive into our methods and results, please check out the paper and code.

📄 LLM Kernel: an evaluation framework for open-ended scientific interpretation

💻 https://github.com/transcriptabio/llm-kernel

This summer we hosted Kalea-Meali’i Try, a high school student at Palo Alto High School. Kalea worked with the computational team to build a tool that groups pathways into functional groups based on their descriptions. Pathways are groups of genes that work together to conduct a variety of biological processes. The Molecular Signatures Database (mSigDB) contains curated sets of pathways that are frequently used to interpret gene expression data and translate gene-level responses into biological process level responses.

However, pathway analysis results pose a significant interpretive challenge. While these gene sets are derived to minimize gene overlaps there is still significant functional overlap between these sets. This leads to significant redundancy in results making it difficult to interpret for the user. At Transcripta, we regularly grapple with the challenge of interpreting pathway responses of 1000s of pathways across 100s of compounds and growing. To address this problem we used Gemini to generate mechanistic descriptions of pathways and used natural language processing (NLP) to group these descriptions into a much smaller set of higher-level functional categories making interpretation that much easier.

Generation of pathway descriptions with Google Gemini

We started by generating pathway descriptions for Reactome, WikiPathways, and Biocarta pathways (total 2800 pathways). These three sets were chosen as they encompass major signaling and metabolic functional annotations. For each pathway, Gemini was queried using Google’s API with the ‘gemini-2.0-flash’ model. The model was provided the pathway name and description (from the respective database) as priors.

One might wonder as to the need of LLM-generated descriptions if descriptions for these pathways are already available. A key reason behind this was to generate coherent and harmonized descriptions because each database might provide slightly different styles of descriptions. For example, Reactome descriptions tend to be entirely mechanistic, like gene X activates gene Y etc. while those from Biocarta might be entire prose style abstracts from a publication. These differences can then affect downstream analysis and skew results.

To highlight the importance of coherent and harmonized descriptions generated using Gemini we used a pre-trained sentence transformer model to embed both raw and LLM-generated descriptions in a reduced vector space. This model is based on BioBERT, a domain-specific language representation model based on BERT (Bidirectional Encoder Representations from Transformers), pre-trained on large-scale biomedical corpora including PubMed abstracts and PMC full-text articles making it better suited for biomedical texts. Embedding text into a vector space converts complex textual data into numerical matrices. In essence, it translates language into numbers so that algorithms can recognize patterns, similarities, and relationships in text the way humans understand context. They most importantly also enable visualization, like the ones shown in figure 1, panels A-B in a 2D space.

Visual inspection of these two panels show that descriptions generated using an LLM form better cluster structure, i.e. groups of points are more distinct and separated from other groups. This observation can be validated using mathematical frameworks to evaluate cluster quality such as Silhouette scores (reflecting intra-cluster cohesion) and Calinski-Harabasz index (reflecting inter-cluster separation) (Fig. 1C-D) both showing superior performance with LLM-generated descriptions.

Grouping pathways using LLM-generated mechanistic descriptions

We used embeddings of LLM descriptions to group pathways into functional groups using topic modeling. Topic modeling is applied to text data to automatically uncover themes in this data, for example to categorize movie synopsis into genres. It can, in this case, be used to organize and summarize complex pathway descriptions by grouping pathways that share similar functions, mechanisms, or underlying biology.

~85% of the 2,800 pathways were successfully grouped into <100 groups (Fig. 2A) of 25-70 pathways. We wanted to make sure these pathway groups weren’t simply based on gene set similarity, such that each pathway group was just made of pathways that shared lots of genes. To do this we computed the pairwise Jaccard Similarity of genes for pathways in each group and found that the mean similarity for groups was <0.5 showing that gene similarity does not by itself account for information contained in each group (Fig. 2B). This is good news because it means these groups are not simply based on similar gene sets but more so on broader similarity encoded in the biological function.

This is also demonstrated by individually examining groups and the function of each pathway in the group. For example, one of the groups corresponds to a cluster of 42 pathways associated with MAPK signaling pathways.

Biological applications of pathway groups

Much like pathway analysis that uses groups of genes to infer response for each pathway, these pathway groups can be used in a similar analysis except for sets of pathways instead of genes to estimate response for each pathway group.

To illustrate this, we randomly picked a CDK4/Cyclin D1, CDK2/Cyclin E and Protein Kinase C (PKC) Inhibitor with a reported PKC IC50 of 0.12 µM from the Novartis MoA Box dataset. CDK2/CDK4 plays an important role in cell cycle maintenance and its inhibition leads to cell cycle arrest. Panels in figure 4 show enrichment (response>0) or de-enrichment (response<0) of these pathway groups. The x-axis shows the magnitude of response measured by a “normalized enrichment score” and the y-axis shows significance (values > 1 are significant).

Interpreting results of analysis of pathway groups was much simpler since redundancy in these groups is minimal and the user is required to only interpret a small number of significantly regulated functional groups before drilling down into detailed per pathway results. Moreover, grouping pathways in this manner also reduces noise in the signal and only reveals coherent functional activity hubs whereas per-pathway results are generally noisier.

Our results quickly revealed that a group of pathways associated with CDKs (“CDK CYCLIN CDC RB”) were down-regulated for 3 out of the 4 concentrations (Fig. 4), as would be expected for a compound annotated as a CDK2/CDK4 inhibitor. We also observed enrichment of a signaling pathway group associated with “NTRK EGR NGF TRKA”. This is an interesting observation because binding of the protein NGF to the receptor TRKA increases EGR expression and activates both PKC and CDK2/CDK4 signaling cascades. The activation of this group of pathways potentially points to a compensatory mechanism in response to the inhibition of PKC/CDK2/CDK4.

Finally we observed a cluster of other signaling pathway groups (labeled in figure 4) that were also activated across 3 out of the 4 concentrations. Interestingly, these pathway groups clustered closely together, based on topic similarity, with the group of NTRK/EGR/NGF/TRKA pathways (Fig. 5).

This provided an additional level of validation that these pathway groups mimic biological function where similar groups of pathways (predicted by topic similarity) are also co-activated during drug treatment.

Conclusions

This is one of the ways in which Transcripta is integrating LLMs with omics data to power discovery and interpretation of cellular function. While LLMs are directly useful and widely used for interpretation of scientific data, they are also becoming very useful for us to power downstream analytical tools such as this one.

This project specifically demonstrated the efficacy of using LLMs to enhance the interpretability of pathway analysis results. By generating coherent and harmonized mechanistic descriptions, and subsequently employing natural language processing and topic modeling, pathways can be grouped into functionally relevant categories. This approach significantly reduces redundancy, simplifies interpretation for users, and reveals coherent functional activity hubs, as evidenced by improved cluster quality metrics and biological validation through drug treatment experiments. These findings suggest a useful new method for analyzing complex gene expression data, offering a clearer and more efficient path to understanding biological processes at a functional level.

Discovering new and viable therapies for genetic diseases is a time-consuming and cost-intensive process, often taking more than a decade and billions of dollars to bring a single new drug to market. For conditions that affect smaller patient populations, the incentives for research and development in the traditional drug discovery lifecycle are often lacking, leaving many patients without hope for a treatment. This reality creates an urgent need for more scalable and efficient solutions. At Transcripta Bio, we’re pioneering a platform that sidesteps the lengthy development of new molecules. By using a high-throughput drug discovery engine, we can uncover how existing, approved drugs impact disease on a cellular level, dramatically accelerating the path to viable treatments.

From a Genetic Anomaly to a Cellular Signature

Our platform’s power is well illustrated by a recent case. We worked with a family whose child was diagnosed with a rare neurodevelopmental disorder, a form of autism spectrum disorder (ASD) linked to chromosomal variants of 19q12. Advanced genetic testing pinpointed the precise cause: a complex genomic rearrangement leading to deficiencies in two critical genes, TSHZ3 and ZNF536, both of which are essential for normal brain development.

To understand the functional impact of this genetic change, we took a small sample of skin cells from the patient and reprogrammed them into neurons in our lab, creating a personalized disease model in a dish. By analyzing the complete set of gene expression patterns, the transcriptome, we gained a comprehensive, 360-degree view of the biological impact of the mutation. This analysis allowed us to identify a clear transcriptomic “disease signature,” a unique fingerprint of cellular dysfunction. In this case, the patient’s neurons showed distinct signs of hyperexcitability, a state where neurons are overly active and fire erratically. This signature became our specific, measurable therapeutic target.

Matching a Disease to a Drug

Our core technology is a platform built on two massive, interconnected datasets: a Disease Signature Atlas containing the transcriptomic fingerprints of various diseases, and a Drug-Gene Atlas detailing how thousands of compounds affect gene expression. By cross-referencing these atlases, our goal is to find drugs that can effectively “reverse” or neutralize a specific disease signature.

We screened a library of compounds against healthy neurons to map how each one altered gene expression. This computational search allowed us to identify a drug with a transcriptomic profile that was a near-perfect mirror image of the patient’s disease signature. The screen identified a promising candidate: entrectinib, a drug already approved for cancer therapy. This was a critical finding, as repurposing an existing drug can significantly shorten the timeline to clinical use. When we applied entrectinib to our lab-grown patient neurons, it successfully and consistently reversed the hyperexcitability signature

A Personalized Trial and a Promising Outcome

With this compelling preclinical data, and in close partnership with the patient’s physician , an “n-of-1” clinical study was initiated. The patient began treatment with a carefully calibrated low dose of entrectinib under close medical monitoring for safety and efficacy measures.

The results have been remarkable. We monitored the patient’s progress by analyzing the transcriptome of whole blood samples collected at regular intervals, which revealed a significant normalization of the disease-related pathways after treatment began. This molecular evidence confirmed the drug was having the intended effect mirroring findings from in vitro neuron treatment. More importantly, these cellular changes translated into real-world benefits, as the patient has shown significant clinical improvements. Standardized developmental assessments, including Vineland scores, have revealed marked improvement in communication and daily living skills, providing a quantitative measure of the treatment’s positive impact.

Looking to the Future

This case is more than just a single success story; it’s a blueprint for a new way of discovering therapies for genetically-defined diseases. We’ve demonstrated that by deeply understanding the molecular basis of a disease, we can use transcriptomics to rapidly match patients with potentially life-changing treatments. This platform-based approach is adaptable and scalable.

We are already extending this discovery pipeline to other complex disease areas and exploring the underlying mechanisms of how these therapies work. Our work with entrectinib suggests that it helps rebalance synaptic activity, offering potential avenues for treating other neurological disorders characterized by hyperexcitability.

We are immensely grateful to the family, their physicians, and the philanthropic partners who made this research possible. Their collaboration was essential to this breakthrough. Together, we are paving the way for a future where every patient with a genetically-defined disease has a clear and rapid path to a personalized therapy.

Palo Alto, CA – May 28, 2025 - Transcripta Bio today announced a research collaboration with Microsoft Research to accelerate the discovery of disease-gene associations using advanced artificial intelligence (AI) and large-scale transcriptomic data. This collaboration brings together Transcripta Bio’s proprietary Drug-Gene Atlas, which contains over a billion gene responses and full-transcriptome data, and Microsoft Research’s machine learning expertise with the goal of identifying novel therapeutic opportunities faster and with more certainty than conventional methods.

Transcripta Bio and Microsoft Research will analyze chemotranscriptomic datasets with a primary goal to identify candidate gene-disease associations supported by experimental evidence of gene expression modulation. The teams hope to accelerate the path from discovery to clinical candidate selection, with the potential for fast-tracking and delivering new therapies to patients more efficiently.

Scott Saponas, Senior Director and Deputy Lab Director, Microsoft Research Health Futures, stated: “We are excited to collaborate with Transcripta Bio to leverage our expertise in rare disease genomics and AI, aiming to develop innovative methods for drug discovery that can significantly impact the treatment of rare diseases.”

Dr. Chris Moxham, Co-founder and CEO of Transcripta Bio, commented: “This collaboration with Microsoft Research enables us to scale our transcriptomics-driven approach, leveraging Microsoft’s computational resources and AI expertise. Together, we are setting a new standard for the speed, certainty, and efficiency of drug discovery.”

About Transcripta Bio

Transcripta Bio is an AI drug discovery company specializing in transcriptomics-driven insights to identify and develop novel therapeutics across multiple disease areas. Using its proprietary AI platform, Transcripta Bio accelerates the discovery and validation of promising drug candidates, leveraging both repurposed and novel molecules for targeted therapeutic interventions.
Website: www.transcriptabio.com
Media Contact: press@transcriptabio.com

Eigenfish, depicted above, are mathematically generated images that we find reflect the innovative qualities of our scientific and computational approach to drug discovery. They have an organic quality that also evokes a sense of order and precision. Each individual Eigenfish is unique but they exist as part of a large family that share a resemblance and aesthetic.

Mathematically, Eigenfish are distributions of eigenvalues in the complex plane for a family of random matrices—in other words, a probabilistic window that reveals deep structure. We find analogies in this with how our platform reveals deep structure in cellular behavior, often using probabilistic methods.

Rather than selecting a single Eigenfish image as emblematic, we chose to embrace their full diversity, generating new ones on-the-fly to provide the visual fingerprint of Transcripta Bio when and where needed. In this way, Eigenfish embody the ethos of Charles Darwin’s famous quote: “from so simple a beginning endless forms most beautiful and most wonderful have been, and are being, evolved.”

Acknowledgement to Simone Conradi

We’d like to graciously thank Professor Simone Conradi for his open source implementation of Eigenfish generation, which was the inspiration and starting point for our adaptation, and his generosity in allowing us to use his work to re-imagine the visual language of Transcripta Bio. We encourage you to explore his other work as well:

• https://profconradi.github.io/

• https://github.com/profConradi

• https://x.com/S_Conradi

Chemo-transcriptomics at scale

The transcriptome occupies a fundamental role as a bedrock biological modality and a vital link in the chain of information of any biological system. At Transcripta Bio, we leverage an omics-first drug discovery approach through the lens of gene expression. Utilizing deep sequencing, scientists capture a comprehensive snapshot of the ongoing dynamic regulatory processes that arise in response to stimuli. These expression quantities unlock several key discovery components, allowing us to simultaneously identify novel targets, uncover enriched disease pathways, and gain insights into regulatory networks. With the wealth of functional knowledge available to us, the hundreds of thousands of data points generated from a single experiment can be leveraged to derive novel insights into the roles that individual genes play in diseases-related biological processes.

Transcripta’s rapid-screening Discovery platform performs high-throughput RNA-Seq assays to screen thousands of unique molecules and quantify their individual effects on a variety of cell types. The resulting millions of perturbations then become training data for our proprietary Conductor AI suite, which predicts gene expression directly from the molecular structure of novel uncharacterized compounds. By unleashing Conductor on massive unlabeled chemical spaces, we are able to discover novel chemo-transcriptomic relationships that:

• Discover new molecular mechanisms that demonstrate improved efficacy

• Navigate off-target effects to improve safety profiles

• Identify compounds with improved target specificity.

Virtual screening for PDE2A downregulation

In an application of Conductor AI’s predictive capabilities, we demonstrate targeted downregulation of PDE2A. Phosphodiesterase type 2A (PDE2A) regulates mitochondria morphology and apoptotic cell death via local modulation of cAMP/PKA signaling, is predominantly expressed in the forebrain, and is known to affect cognitive processes. Inhibition of PDE2A has been associated with improved behavior symptoms in autism spectrum disorder and better stroke recovery via increased neuroregeneration.

With conventional drug-discovery approaches, targeting a specific protein for inhibition is often a tedious and unpredictable endeavor, typically yielding limited success in identifying molecules with the desired specificity and efficacy. In the case of PDE2A, this challenge is exacerbated due to the presence of numerous other PDE genes, making selectivity crucial to avoid interfering with the remaining phosphodiesterases. Our Conductor AI platform, trained on our extensive proprietary dataset, navigates this barrier by predicting the high-dimensional transcriptomic response of any given molecular structure, including 23 different PDE genes. This allows us to perform high-throughput virtual screening of billions of compounds using massive chemical catalogs, enabling the discovery of small molecules that are potent and specific PDE2A downregulators.

In silico activity and downstream experimental validation

Utilizing a billion-scale chemical catalog of novel synthesizable compounds as a search space, Conductor identified a small cohort of molecules that were predicted to induce inhibitory effects on PDE2A. Following the synthesis of these putative actives, in vitro validation using bulk RNA-seq in glutamatergic neurons confirmed their efficacy, demonstrating dose-dependent PDE2A downregulation. Concentration-response curves for two of these chemical inhibitors, shown below, depict the transcript log-fold change relative to baseline levels, illustrating the inhibitory effect.

Translatability from transcript abundance to protein product is a key criterion for any successful drug program. We find that applying the two compounds significantly decreases protein abundance when quantified with protein immunoblotting.

Our roadmap

Going forward, we are poised to significantly enhance our modeling capabilities and uncover innovative therapeutic opportunities as we continue to scale data generation both in the number of screened molecules and in the cell types we screen. Screening molecules strategically selected to expose therapeutic opportunities in unexplored chemical space will significantly improve the precision and scope of our predictive models. Similarly, expanding the diversity of disease-relevant cell types we screen uncovers opportunities in more therapeutic areas while simultaneously facilitating cross–cell type modeling of transcriptomic activity.

In addition to growing our experiment dataset, we are deeply invested in broadening Transcripta’s AI Research Program to derive new quantitative and computational techniques that push the boundaries of biological data modeling in areas such as non-parametric modeling of complex interacting systems, causal modeling of noisy high-dimensional data-generating processes, and multimodal representation learning. Additionally, our experiments with molecular generative models, like NVIDIA’s MolMIM, for molecular property refinement and lead optimization have shown there is promise for profound progress in molecular representation and generation, especially in concert with transcriptomic data.

We look forward to sharing our progress as we continue to grow Conductor’s capability to learn from and predict drug responses across different cellular contexts, broadening the therapeutic areas that we can address and furthering our goal of rapid drug discovery for transformative patient impact.

At Transcripta, we are charting a faster path in drug discovery to create better lives for people around the world. If this interests you and you’d like to collaborate, please contact partnerships@transcriptabio.com.

‍

Palo Alto, Calif., May 30, 2024 – Transcripta Bio, a platform drug development biotechnology company that is revolutionizing discovering and developing medicines at unprecedented speed, scale, and capital efficiency, has been included in the fourth-annual TIME100 Most Influential Companies list. The prestigious list spotlights companies around the world that are making an extraordinary impact and recognizes Transcripta Bio’s significant achievements in speeding drug discovery.

“We are honored to be included in the TIME100 Most Influential Companies list and recognized for our groundbreaking advancements in drug discovery,” said Chris Moxham, co-founder, CEO, and Chief Scientific Officer, Transcripta Bio. “‘Charting a faster path in drug discovery to create better lives for people around the world’ is our mission and having our strong progress towards achieving this acknowledged today is a milestone moment. I want to thank our world-class team of scientists and technology leaders in AI and ML for their outstanding work and dedication to helping humanity.”

The TIME100 Most Influential Companies list acknowledges Transcripta Bio for “dramatically reducing the time and money it takes to discover drugs —not just for rare diseases, but for any drug that treats disease by changing gene expression. Transcripta Bio’s Drug-Gene Atlas database shows how small molecules affect gene expression, allowing it to identify new uses for already-approved pharmaceuticals and, with the help of AI, novel candidates. In less than two years, Transcripta moved five FDA-approved drugs into single-patient studies for new indications, which otherwise might have taken five to eight years.”

Transcripta Bio is dedicated to high certainty drug discovery. By focusing on diseases where the causality is unequivocally established by human genetics, Transcripta Bio leverages the universe of small molecules that has consistently led to successful drugs. Transcripta Bio’s Drug-Gene Atlas maps molecular structure and drug action to the whole cell transcriptome, revealing therapeutic candidates that rebalance gene expression to treat disease. The proprietary artificial intelligence modeling suite, Conductor AI, utilizes the Drug-Gene Atlas to discover and predict the effects of new drugs at transcriptome scale. By building cutting-edge technology and forging strategic partnerships, Transcripta Bio is dramatically accelerating the global drug discovery process.

To assemble the list, TIME solicited nominations across sectors, and polled its global network of contributors and correspondents, as well as outside experts. TIME editors evaluated each on key factors, including impact, innovation, ambition, and success. The result is a diverse group of 100 businesses helping chart an essential path forward. Read the full list here: time.com/100companies.

About Transcripta Bio

Transcripta Bio is dedicated to high-certainty drug discovery, revolutionizing the process of discovering and developing medicines at unprecedented speed, scale, and capital efficiency.

With data from hundreds of millions of experiments, we have mapped the effects of thousands of compounds on gene expression to discover new therapies. Using technology at scale, we discover new therapies for multiple diseases simultaneously. Transcripta is headquartered in Palo Alto, Calif. For more information, please visit www.transcriptabio.com/.

Today we are unveiling Transcripta Bio.

After working tirelessly for the last two years and completing more than 200 million experiments in our labs, we are opening the aperture of drug discovery to impact countless patients and lives.

We have already seen rapid proof of our platform from the early results across a number of rare diseases for which there are no current treatments. This will remain an important area of focus for us. However, there is so much more that we can do — hence our company’s new name.

At Transcripta, we are dedicated to high certainty drug discovery. By focusing on diseases where the causality is unequivocally established by human genetics, we’re leveraging the universe of small molecules that has consistently led to successful drugs and using technology at scale to discover new therapies for many diseases all at once. We also believe that the transcriptome holds immense potential for innovative discovery, and thus it is the cornerstone of our platform focus. Not only is the transcriptome the Rosetta Stone that describes a given cell state under healthy or disease settings, but it is also a gateway to characterize the fundamental basis of drug action and to discover new medicines.

We continue to develop the premier Drug-Gene Atlas in our high throughput lab that maps molecular structure and drug action to the whole cell transcriptome, revealing therapeutic candidates that rebalance gene expression to treat disease. Those “wet lab” results are also fed into our proprietary artificial intelligence model, Conductor AI, to predict new drugs that move gene expression as appropriate. By mapping structure to function, we are learning the essential “chemistry rules” to tune gene expression. From the machine learning models we have trained on thousands of genes, we are virtually screening billions of compounds for disease-modifying ability while minimizing off-target effects. In essence, we are building a closed-loop drug machine.

There has been so much support for Transcripta given our strong progress that we received pre-emptive interest from investors in raising a capital round. Given this, we just closed a $10 million round from current investors, including JAZZ Venture Partners and BlueYard Capital, as well as participation from family offices with personal interest and excitement about the promise of our platform. This capital will allow us to continue to accelerate and scale our work.

In this next chapter, we are also partnering with leading biopharma and tech companies, research centers, and patient foundations to uncover and advance new opportunities. As we chart a faster and broader path in drug discovery to create better lives for people around the world, we hope you’ll join us on this journey.

Chris Moxham, PhD
Co-founder, CEO, Chief Scientific Officer

See full press release on PR Newswire.

Palo Alto, Calif., October 4 -- Rarebase, a platform drug development biotechnology company that is revolutionizing discovering and developing medicines at unprecedented speed, scale, and capital efficiency, today announced that Chris Moxham, Ph.D., President and Chief Scientific Officer of Rarebase, has been named Chief Executive Officer effective immediately. Moxham assumes the CEO position from Rarebase co-founder Onno Faber, who has been appointed to the newly created role of Chief Technology Officer.

"It is a privilege to lead Rarebase in this next phase of growth," Moxham said. "We are on the path to dramatically accelerate the way the world does drug discovery. We've proven our platform works and we have already uncovered potential treatments for hundreds of rare diseases. Our next phase will see our 'drug-gene atlas' enable AI-driven drug discovery at scale and we'll grow our world-class team further as we work tirelessly to serve the needs of everyone living with a rare disease."

Rarebase is focused on improving the lives of 400 million people living with a rare disease, 95% of whom don't have an effective treatment. With a product engine designed to discover treatments for hundreds of monogenic diseases simultaneously, Rarebase is squarely focused on the largest untapped drug discovery opportunity in human health: rare diseases.

"Chris is an exceptional, deeply committed leader who is highly regarded for his strategic vision and operational excellence," said John Lee, Partner, JAZZ Venture Partners. "We couldn't be more excited to have Chris at the helm as a new chapter begins. Now is the time for Rarebase to accelerate its growth and leverage the potential treatments it has already uncovered for hundreds of rare diseases."

Added Lee, "We are also pleased to have Onno assume the new role of Chief Technology Officer and are grateful for his unwavering passion for serving the enormous unmet need for patients living with a rare disease."

Moxham joined Rarebase in November, 2021, as Chief Scientific Officer and was elevated to President in July, 2023. His career spans more than 25 years in drug discovery, particularly cutting-edge science and drug hunting in large pharma and biotech.

Moxham spent two decades at Eli Lilly and Co., most recently as the Vice President of Quantitative Biology, where he helped bring more than 10 molecules into the clinic across multiple therapeutic areas and across both small and large molecules.

He then served as Chief Scientific Officer at Fulcrum Therapeutics, a clinical stage biotech focused on rare diseases, where he was responsible for the full range of drug discovery, translational science and early development activities, including building and advancing the company's preclinical portfolio through Phase 1 clinical trials. He was the scientific lead for FTX-6058, which is currently in phase 1 clinical testing for Sickle Cell Disease. Moxham was instrumental in creating multiple business development partnerships and with engaging the investment community to achieve multiple rounds of financing for the company.

About Rarebase, PBC

Rarebase is a true platform drug development biotech that is revolutionizing the process of discovering and developing medicines at unprecedented speed, scale, and capital efficiency.

Rarebase is focused on enabling a better life for the 400 million people living with rare disease. With a product engine designed to discover treatments for hundreds of monogenic diseases simultaneously, Rarebase is squarely focused on the largest untapped drug discovery opportunity in human health: rare diseases. Rasebase has already uncovered potential treatments for hundreds of rare diseases. The company has also built a "drug-gene atlas" to enable AI-driven drug discovery. Rarebase is headquartered in Palo Alto, Calif. For more information, please visit www.rarebase.com.