Event
AlphaFold sagt Proteinstrukturen voraus
On 15 July 2021 a research paper described AlphaFold, a neural network for protein structure prediction. The authors evaluated it in part using the CASP14 competition. Predictions are not automatically experimental confirmation. Source
A week later, researchers described predictions for 98.5% of proteins in the human reference proteome. Local confidence reached the selected confidence range for 58% of amino-acid residues and was very high for 36%. These percentages have different denominators. On 22 July DeepMind and EMBL announced a freely accessible database. Estimated confidence can vary within a single protein. Source
Sources
Nature-Forschungsarbeit: Veröffentlichung, Methode und Auswertung.
Proteomstudie und EMBL-Meldung: Abdeckung, Konfidenz und Datenbank.
Offen
AlphaFold nicht mit dem Go-System AlphaGo gleichsetzen; Vorhersagequalität hängt von der jeweiligen Struktur ab. / Do not conflate AlphaFold with the Go system AlphaGo; prediction quality varies by structure.