Protein design aims to build novel proteins customized for specific purposes, thereby holding the potential to tackle many environmental and biomedical problems. Recent progress in Transformer-based architectures has enabled the implementation of language models capable of generating text with human-like capabilities. Here, motivated by this success, we describe ProtGPT2, a language model trained on the protein space that generates de novo protein sequences following the principles of natural ones. The generated proteins display natural amino acid propensities, while disorder predictions indicate that 88% of ProtGPT2-generated proteins are globular, in line with natural sequences. Sensitive sequence searches in protein databases show that ProtGPT2 sequences are distantly related to natural ones, and similarity networks further demonstrate that ProtGPT2 is sampling unexplored regions of protein space. AlphaFold prediction of ProtGPT2-sequences yields well-folded non-idealized structures with embodiments and large loops and reveals topologies not captured in current structure databases. ProtGPT2 generates sequences in a matter of seconds and is freely available.

Purpose: Patients with colorectal cancer who respond to the anti-EGFR antibody cetuximab often develop resistance within several months of initiating therapy. To design new lines of treatment, the molecular landscape of resistant tumors must be ascertained. We investigated the role of mutations in the EGFR signaling axis on the acquisition of resistance to cetuximab in patients and cellular models. Experimental Design: Tissue samples were obtained from 37 patients with colorectal cancer who became refractory to cetuximab. Colorectal cancer cells sensitive to cetuximab were treated until resistant derivatives emerged. Mutational profiling of biopsies and cell lines was performed. Structural modeling and functional analyses were performed to causally associate the alleles to resistance. Results: The genetic profile of tumor specimens obtained after cetuximab treatment revealed the emergence of a complex pattern of mutations in EGFR, KRAS, NRAS, BRAF, and PIK3CA genes, including two novel EGFR ectodomain mutations (R451C and K467T). Mutational profiling of cetuximab-resistant cells recapitulated the molecular landscape observed in clinical samples and revealed three additional EGFR alleles: S464L, G465R, and I491M. Structurally, these mutations are located in the cetuximab-binding region, except for the R451C mutant. Functionally, EGFR ectodomain mutations prevent binding to cetuximab but a subset is permissive for interaction with panitumumab. Conclusions: Colorectal tumors evade EGFR blockade by constitutive activation of downstream signaling effectors and through mutations affecting receptor–antibody binding. Both mechanisms of resistance may occur concomitantly. Our data have implications for designing additional lines of therapy for patients with colorectal cancer who relapse upon treatment with anti-EGFR antibodies. Clin Cancer Res; 21(9); 2157–66. ©2015 AACR.

The experimental characterization and computational prediction of protein structures has become increasingly rapid and precise. However, the analysis of protein structures often requires researchers to use several software packages or web servers, which complicates matters. To provide long-established structural analyses in a modern, easy-to-use interface, we implemented ProteinTools, a web server toolkit for protein structure analysis. ProteinTools gathers four applications so far, namely the identification of hydrophobic clusters, hydrogen bond networks, salt bridges, and contact maps. In all cases, the input data is a PDB identifier or an uploaded structure, whereas the output is an interactive dynamic web interface. Thanks to the modular nature of ProteinTools, the addition of new applications will become an easy task. Given the current need to have these tools in a single, fast, and interpretable interface, we believe that ProteinTools will become an essential toolkit for the wider protein research community. The web server is available at https://proteintools.uni-bayreuth.de.

The design of functional enzymes holds promise for transformative solutions across various domains but presents significant challenges. Inspired by the success of language models in generating nature-like proteins, we explored the potential of an enzyme-specific language model in designing catalytically active artificial enzymes. Here, we introduce ZymCTRL (’enzyme control’), a conditional language model trained on the enzyme sequence space, capable of generating enzymes based on user-defined specifications. Experimental validation at diverse data regimes and for different enzyme families demonstrated ZymCTRL’s ability to generate active enzymes across various sequence identity ranges. Specifically, we describe the design of carbonic anhydrases and lactate dehydrogenases in zero-shot, without requiring further training of the model, and showcasing activity at sequence identities below 40% compared to natural proteins. Biophysical analysis confirmed the globularity and well-folded nature of the generated sequences. Furthermore, fine-tuning the model enabled the generation of lactate dehydrogenases outside of natural sequence space but with activity comparable to their natural counterparts. Two of the artificial lactate dehydrogenases were selected for scale production and successfully lyophilised, maintaining activity and demonstrating preliminary conversion in one-pot enzymatic cascades under extreme conditions. Our findings open a new door towards the rapid and cost-effective design of artificial proficient enzymes. The model and dataset are freely available to the community.