Researchers from the University of Münster have developed an evolutionary algorithm that generates customized “molecular fingerprints” to predict molecular properties and reactions accurately. This method mimics natural selection, using mutation and recombination to find optimal molecular representations. It aids in predicting quantum chemical properties and toxicity without requiring expert knowledge, reducing human bias. While other methods may excel with extensive data or expert input, this approach is versatile and interpretable, helping understand why specific molecular structures influence predictions.
For more details, visit the full article here: