The widespread adoption of high-throughput omics technologies has exponentially increased the amount of protein sequence data involved in many salient disease pathways and their respective therapeutics and diagnostics. Despite the availability of large-scale sequence data, the lack of experimental fitness annotations underpins the need for self-supervised and unsupervised machine learning (ML) methods. These techniques leverage the meaningful features encoded in abundant unlabeled sequences to accomplish complex protein engineering tasks. Proficiency in the rapidly evolving fields of protein engineering and generative AI is required to realize the full potential of ML models as a tool for protein fitness landscape navigation. Here, we support this work by (i) providing an overview of the architecture and mathematical details of the most successful ML models applicable to sequence data (e.g. variational autoencoders, autoregressive models, generative adversarial neural networks, and diffusion models), (ii) guiding how to effectively implement these models on protein sequence data to predict fitness or generate high-fitness sequences and (iii) highlighting several successful studies that implement these techniques in protein engineering (from paratope regions and subcellular localization prediction to high-fitness sequences and protein design rules generation). By providing a comprehensive survey of model details, novel architecture developments, comparisons of model applications, and current challenges, this study intends to provide structured guidance and robust framework for delivering a prospective outlook in the ML-driven protein engineering field.

Discovering new binding function via a combinatorial library in small protein scaffolds requires balance between appropriate mutations to introduce favorable intermolecular interactions while maintaining intramolecular integrity. Sitewise constraints exist in a non-spatial gradient from diverse to conserved in evolved antibody repertoires; yet non-antibody scaffolds generally do not implement this strategy in combinatorial libraries. Despite the fact that biased amino acid distributions, typically elevated in tyrosine, serine, and glycine, have gained wider use in synthetic scaffolds, these distributions are still predominantly applied uniformly to diversified sites. While select sites in fibronectin domains and DARPins have shown benefit from sitewise designs, they have not been deeply evaluated. Inspired by this disparity between diversity distributions in natural libraries and synthetic scaffold libraries, we hypothesized that binders resulting from discovery and evolution would exhibit a non-spatial, sitewise gradient of amino acid diversity. To identify sitewise diversities consistent with efficient evolution in the context of a hydrophilic fibronectin domain, >10⁵ binders to six targets were evolved and sequenced. Evolutionarily favorable amino acid distributions at 25 sites reveal Shannon entropies (range: 0.3–3.9; median: 2.1; standard deviation: 1.1) supporting the diversity gradient hypothesis. Sitewise constraints in evolved sequences are consistent with complementarity, stability, and consensus biases. Implementation of sitewise constrained diversity enables direct selection of nanomolar affinity binders validating an efficient strategy to balance inter- and intra-molecular interaction demands at each site.

Background: Extracellular vesicles (EVs) are attracting interest as a new class of drug delivery vehicles due to their intrinsic nature of biomolecular transport in the body. We previously demonstrated that EV surface modification with tissue-specific molecules accomplished targeted EV-mediated DNA delivery. Methods: Here, we describe reliable methods for (i) generating EGFR tumor-targeting EVs via the display of high-affinity monobodies and (ii) in vitro measurement of EV binding using fluorescence and bioluminescence labeling. Monobodies are a well-suited class of small (10 kDa) non-antibody scaffolds derived from the human fibronectin type III (FN3) domain. Results: The recombinant protein consists of the EGFR-targeting monobody fused to the EV-binding domain of lactadherin (C1C2), enabling the monobody displayed on the surface of the EVs. In addition, the use of bioluminescence or fluorescence molecules on the EV surface allows for the assessment of EV binding to the target cells. Conclusions: In this paper, we describe methods of EV engineering to generate targeted delivery vehicles using monobodies that will have diverse applications to furnish future EV therapeutic development, including qualitative and quantitative in vitro evaluation for their binding capacity.