In-vitro laboratory endpoints can be combined with computational predictive models and synthetic human tissues to assess clinical toxicity

Overview Debates Related claims Topics

Traditional drug development relies on empirical trial-and-error screening processes rather than predictive computational modeling

Isomorphic Labs trains computational models to predict molecular structures and interactions at experimental accuracy

Isomorphic Labs partners with biopharmaceutical companies to apply its artificial intelligence drug design engine to partner drug targets

Isomorphic Labs was created with the stated mission of solving all human disease

AlphaFold's demonstration of protein structure prediction served as the proof of concept that led to the founding of Isomorphic Labs

Preclinical toxicity studies conducted in rats are approximately 30 percent predictive of clinical toxicity in humans

Most candidate drugs fail during human clinical trials rather than during preclinical in-vitro research

Isomorphic Labs utilizes automated laboratories to generate experimental data to power computational models

Drug regulatory frameworks historically mandated testing candidate molecules on rats and other animals before human clinical trials

Patients facing late-stage terminal cancers typically possess a higher tolerance for treatment risk than patients treated for chronic low-level inflammation

AlphaGenome increased genomic prediction resolution from Enformer's 128 base pairs down to single base-pair resolution

AlphaGenome Atlas precomputed variant effect predictions across all 9 billion possible single-nucleotide mutations in the human genome

In 2025, Google DeepMind unveiled AlphaGenome to predict the function of non-coding DNA across whole tissues

AlphaGenome 2 allows external researchers to input custom cell-type data to generate variant effect predictions without fine-tuning the model

AlphaGenome 2 evaluates variant impacts at the individual cell-type level rather than aggregate bulk tissue measurements

AlphaGenome 2 predicted that a disease-associated non-coding variant regulates a target gene located 290,000 base pairs away

Demis Hassabis proposed the virtual cell vision as a computational system simulating cellular responses to genetic alterations and drug interventions

Enformer was developed by Google DeepMind in collaboration with Calico to read DNA sequences

Broad Institute researcher Laura Kovel used the AlphaGenome Variant Impact score to identify the causal splicing variant in an infant with pediatric epilepsy

AlphaGenome expanded input sequence length from Enformer's 200,000 base pairs to 1,000,000 base pairs