Guest Column | November 16, 2026

Open-Source Model Predicts Huntington's Drug Performance

A conversation between P. Peter Ghoroghchian, MD, Ph.D., at Latus Bio and Life Science Connect's Jon O'Connell

Looking-GettyImages-1207121867

Huntington's disease is driven by the cumulative expansion of a CAG trinucleotide repeat, and the field now has a credible way to slow that expansion. What it has lacked is a quantitative answer for critical questions that help direct treatment, including how long before a patient notices a difference.

Latus Bio built a computational model to answer it. Drawing on natural history data from the Enroll-HD database, the model translates degrees of MSH3 reduction and neuronal coverage into predicted changes in clinical trajectory — a 50% to 120% slowing of progression on the composite Unified Huntington Disease Rating Scale (cUHDRS) at the upper end of its simulations. The company published the work in Nature Communications earlier this year, then released the model itself, modality-agnostic and free to anyone pursuing somatic instability by any route.

In part one of this conversation, Latus CEO P. Peter Ghoroghchian, MD, Ph.D., walks through what the model predicts, what its assumptions rest on, and why Latus is giving it away for free.

In this study, the model and vector play equally important parts. Can you start by describing how both inform each other?

This is Part 1 of an article exploring Latus Bio's model for predicting CAG repeat suppression in Huntington's disease therapy. Part 2 discusses Latus's preclinical strategy toward IND.

Click here to read Part 2.

Ghoroghchian: Our approach brings together computational modeling and experimental gene therapy development. The model uses human data to predict how reducing somatic CAG repeat expansion could translate into meaningful clinical benefits for Huntington’s disease (HD) patients, including delays in symptom onset and slowing disease progression. It helps us understand the degrees of MSH3 reduction and coverage of vulnerable neuronal populations that are potentially required to achieve these outcomes.

We then test the model-derived predictions experimentally using LTS-201, our investigational AAV gene therapy designed to deliver an engineered microRNA targeting MSH3, to HD-target cells in the deep brain and cerebral cortex. By incorporating our preclinical findings into the model, we can better inform dose selection, patient populations, and clinical development strategies. This creates a framework in which the computational and experimental approaches continuously inform one another.

The abstract says the modeling predicts 50% to 120% slowing of progression, which we should note is not reversal at the high end. Correct? Can you give us a quick explainer of what that range describes?

Ghoroghchian: That's correct. The 50% to 120% range represents model-predicted differences in clinical progression relative to natural history patients from the Enroll-HD database of Huntington's disease, as measured by changes in the composite Unified Huntington Disease Rating Scale (cUHDRS). These are computational predictions for a defined patient population based on preclinical MSH3 reduction, not observed clinical outcomes for any individual.

Latus Bio released its Huntington's disease model and webapp on GitHub.

Find it here.

Importantly, a result above 100% should not be interpreted as evidence of disease reversal or recovery of neurons already lost. It reflects how the modeled treatment population trajectory compares with the expected trajectory for the untreated patient population over a specified period. The key takeaway is the potential to substantially alter disease progression by slowing the underlying process of somatic CAG repeat expansion. Whether those predictions translate into clinical benefit will need to be established in prospective human studies.

You note the registry data is binned in five-year increments to protect patient privacy, so you cannot establish error rates or variance. How does that limit modeling?

Ghoroghchian: This is an important limitation that we acknowledge. Because the underlying clinical data is aggregated into five-year intervals, we cannot fully capture individual patient variability or calculate certain measures of uncertainty with the precision that patient-level longitudinal data would allow.

However, our model incorporates established relationships between inherited CAG repeat length, age, somatic expansion, and disease progression across the spectrum of HD patients. We also evaluated variability across repeated simulations, which produced consistent results under the same assumptions.

We view the model as a tool for generating testable predictions and informing therapeutic development, rather than a definitive forecast for any individual patient. As additional longitudinal datasets become available, we can further refine and validate its predictive accuracy.

The model rests on expansion rates and thresholds of 150 and 300 repeats. Those are the best available, but they are also assumptions. How are you hedging against the uncertainty?

Ghoroghchian: The thresholds are informed by published research from donated postmortem brains of HD patients, examining how CAG repeat expansion affects function and survival of vulnerable neurons in HD. The CAG repeat threshold at which healthy neurons experience transcriptional dysregulation has been established as 150 CAG repeats; at 300 repeats, the neurons undergo even more rapid expansion, leading to neuronal loss. This biological process, occurring in an asynchronous and stochastic fashion, across entire populations of human neurons is what is captured by our novel model.

We recognize that these are human biologically-informed modeling assumptions rather than absolute boundaries that apply identically to every neuron or patient. That's why we designed the model to allow researchers to adjust expansion rates, thresholds, and other parameters to explore how those changes affect predicted outcomes. Importantly, as new data become available, for example, better estimation of thresholds from human data, the model can accommodate this new information.

Ultimately, the most important question is whether reducing somatic instability translates into meaningful clinical benefit. Our preclinical studies provide evidence that LTS-201 can reduce MSH3 expression and somatic expansion, but clinical studies will be necessary to validate the predicted therapeutic effects.

The modeling consistently favors early intervention and dosing asymptomatic people who carry the mutation. How do you think about risk and benefit? Has the HD community weighed in?

Ghoroghchian: One of the most compelling findings from our model is that intervening earlier could provide the greatest potential benefit, before extensive CAG repeat expansion and neuronal damage have occurred.

Huntington's disease is particularly well suited to exploring intervention in both symptomatic (initial goal) as well as (eventually) in premanifest patients, as genetic testing can identify individuals within families who carry the disease-causing mutation at any time.

However, treating premanifest individuals introduces important ethical and clinical considerations. Any preventive approach would require compelling evidence of safety, demonstrated efficacy in symptomatic patients, a clear understanding of the potential benefits of earlier treatment, and careful consideration of an individual's genetic profile, age, and predicted disease trajectory.

Engagement with patients, families, advocacy organizations, and clinicians will be essential in informing these decisions. We would not want to suggest that the model alone justifies treating premanifest individuals. Rather, it provides a scientific rationale for investigating whether earlier intervention could meaningfully change the course of the disease.

Finally, you put the model, which you say is modality-agnostic, in the public domain. Why did you do that?

Ghoroghchian: Huntington's disease is a devastating condition, and we believe advancing the field requires collaboration and scientific transparency.

Although we developed the model alongside our own gene therapy program, its underlying framework is not specific to LTS-201 or AAV delivery. It can be applied to other therapeutic approaches designed to reduce somatic instability, including small molecules, antisense oligonucleotides, RNA interference, and potentially other platform technologies. By making the model publicly available, we hope to provide researchers with a resource to explore therapeutic hypotheses, evaluate potential treatment thresholds, and better understand how interventions might translate into clinical benefit. We believe this will encourage scientific discussion, facilitate independent evaluation, and, ultimately, help accelerate the development of effective treatments for the Huntington's disease community.

The model is publicly accessible at latushdmodel.com, and the source code is available through GitHub.

About The Expert:

P. Peter Ghoroghchian, MD, Ph.D., is the CEO of Latus Bio. Previous roles include cofounder, president, and CEO of Ceptur Therapeutics; senior vice president and head of therapeutic development at Repertoire Immune Medicines, a Flagship Pioneering company; chief technology officer at Poseida Therapeutics; and cofounder and chief scientific officer at Vindico NanoBioTechnology. He received his MD and Ph.D. in bioengineering from the University of Pennsylvania.