Brain drug development is stuck: more than 90% of drugs that pass animal trials fail in humans, and for Alzheimer's disease that figure reaches 99.6%. Regulators worldwide are demanding a shift to animal-free methods, but organoids alone don't solve the problem. What's needed is a way to turn their complex biological data into clear, comparable ratings. This work stands apart: nearby studies in our archive deal with brain-computer interfaces, predicting surgical outcomes, and MRI analysis, but nobody had proposed a unified scale for drug screening on organoids.
The researchers used a multi-region brain organoid (MRBO) grown from human stem cells and aged to 200 days. It contains regions of the cortex, midbrain, brainstem, and its own blood vessels. The organoids were exposed to five drugs with different mechanisms: a GLP-1 agonist, bupropion, fluoxetine, fingolimod, and an AKT activator. Four types of data were collected: single-nucleus RNA sequencing, bulk sequencing, proteomics, and electrical activity on a multielectrode array. Then the NIS was built: every gene, protein, and electrical parameter was preassigned a "polarity", positive if the change is favorable for the brain, negative if harmful. These polarities were averaged within the four biological categories and summed into a final score.
Fingolimod received the highest total score, 1.06, landing in the "green" zone thanks to positive ratings for development (0.53) and resilience (0.65). GLP-1 scored 0.66 with the highest resilience rating (0.91), but negative scores for inflammation and degeneration. Bupropion (0.42), fluoxetine (0.44), and SC79 (0.27) fell into the "yellow" zone. Comparison with rodent data revealed an almost random match at the level of individual genes: for GLP-1, only 51.6% of genes shifted in the same direction as in mice. At the category level, however, the NIS showed agreement for bupropion and fluoxetine, while for GLP-1 the human organoid gave a positive signal even though the mouse data hovered around zero. This was explained by the vascular component in the organoid, which is absent from rodent neural samples.
What sets this work apart from its neighbors in the archive is that it offers not a new data analysis method but a way to interpret data for decision-making. Nearby studies focus on the accuracy of neural interfaces or image classification, whereas here a scale has been created that applies to any model and any species. It's a first step toward standardizing drug assessment on organoids, which could speed up preclinical trials and reduce reliance on animals.
The NIS can be extended to organoids from patients with specific diseases, to other types of microphysiological systems, and to broader drug panels. Because the polarity dictionary is fixed, new data will be comparable with old data without retraining. Versions of the index for other organs, such as the liver or heart, may emerge.
The method will influence CNS pharmacology, toxicology, and personalized medicine by providing a common language for describing drug effects on neural tissue.
The immediate next steps include testing the NIS on organoids with disease-associated mutations and expanding the drug panel for validation.
Neuroscience still hasn't resolved how to translate molecular changes into functional outcomes. The NIS directly links gene and protein expression to categories of neuronal health, but it doesn't solve the problem of predicting clinical efficacy in an individual patient.
The organoid grew for 200 days, roughly 6.5 months, which is comparable to the second trimester of a human pregnancy. In that time it formed 26 cell types, including cells of the vascular network.
Not published on arXiv. Here are its page, the PDF and the formalization: you can check it yourself, not only read our account.