Machine learning (ML) is becoming an attractive tool in mutagenesis-based protein engineering because of its ability to design a variant library containing proteins with a desired function. However, it remains unclear how ML guides directed evolution in sequence space depending on the composition of training data. Here, we present a ML-guided directed evolution study of an enzyme to investigate the effects of a known “highly positive” variant (i.e., variant known to have high enzyme activity) in training data. We performed two separate series of ML-guided directed evolution of Sortase A with and without a known highly positive variant called 5M in training data. In each series, two rounds of ML were conducted: variants predicted by the initial round were experimentally evaluated and used as additional training data for the second-round of prediction. The improvements in enzyme activity were comparable between the two series, both achieving enzyme activity 2.2–2.5 times higher than 5M. Intriguingly, the sequences of the improved variants were largely different between the two series, indicating that ML guided the directed evolution to the distinct regions of sequence space depending on the presence/absence of the highly positive variant in the training data. This suggests that the sequence diversity of improved variants can be expanded not only by conventional ML using the whole training data but also by ML using a subset of the training data even when it lacks highly positive variants. In summary, this study demonstrates the importance of regulating the composition of training data in ML-guided directed evolution.

We previously reported the use of a humanized bi‐specific diabody that targets epidermal growth factor receptor and CD3 (hEx3‐Db) for cancer immunotherapy. Bacterial expression can be used to express small recombinant antibodies on a large scale; however, their overexpression often results in the formation of insoluble aggregates, and in most cases artificial affinity peptide tags need to be fused to the antibodies for purification by affinity chromatography. Here, we propose a novel method for preparing refined, functional, tag‐free bi‐specific diabodies from IgG‐like bi‐specific antibodies (BsAbs) in a mammalian expression system. We created an IgG‐like BsAb in which bi‐specific diabodies were fused to the human Fc region via a designed human rhinovirus 3C (HRV3C) protease recognition site. The BsAb was purified by protein A affinity chromatography, and the refined tag‐free hEx3‐Db was efficiently produced from the Fc fusion format by protease digestion. The tag‐free hEx3‐Db from the Fc fusion format showed a greater inhibition of cancer growth than affinity‐tagged hEx3‐Db prepared directly from Chinese hamster ovary cells. We also applied our novel method to another small recombinant antibody fragment, hEx3 single‐chain diabody (hEx3‐scDb), and demonstrated the versatility and advantages of our proposed method compared with papain digestion of hEx3‐scDb. This approach may be used for industrial‐scale production of functional tag‐free small therapeutic antibodies.

Despite the advances in surface-display systems for directed evolution, variants with high affinity are not always enriched due to undesirable biases that increase target-unrelated variants during biopanning. Here, our goal was to design a library containing improved variants from the information of the “weakly enriched” library where functional variants were weakly enriched. Deep sequencing for the previous biopanning result, where no functional antibody mimetics were experimentally identified, revealed that weak enrichment was partly due to undesirable biases during phage infection and amplification steps. The clustering analysis of the deep sequencing data from appropriate steps revealed no distinct sequence patterns, but a Bayesian machine learning model trained with the selected deep sequencing data supplied nine clusters with distinct sequence patterns. Phage libraries were designed on the basis of the sequence patterns identified, and four improved variants with target-specific affinity (EC₅₀ = 80–277 nM) were identified by biopanning. The selection and use of deep sequencing data without undesirable bias enabled us to extract the information on prospective variants. In summary, the use of appropriate deep sequencing data and machine learning with the sequence data has the possibility of finding sequence space where functional variants are enriched.

We previously reported the utility for cancer immunotherapy of a humanized bispecific diabody (hEx3) that targets epidermal growth factor receptor and CD3. Here, we used dynamic and static light scattering measurements to show that the multimer fraction observed in hEx3 in solution is a monodisperse tetramer. The multimerization into tetramers increased the inhibition of cancer cell growth by the hEx3 diabody. Furthermore, 1:2 stoichiometric binding for both antigens was observed in a thermodynamic analysis, indicating that the tetramer has bivalent binding activity for each target, and the structure may be in a circular configuration, as is the case for the single-chain Fv tetrabody. In addition to enhanced cytotoxicity, the functional affinity and stability of the hEx3 tetrabody were superior to those of the hEx3 diabody. The increase in molecular weight is also expected to improve the pharmacokinetics of the bispecific diabody, making the hEx3 tetrabody attractive as a therapeutic antibody fragment for cancer immunotherapy.

Small bispecific antibodies that induce T-cell–mediated cytotoxicity have the potential to damage late-stage tumor masses to a clinically relevant degree, but their cytotoxicity is critically dependent on their structural and functional properties. Here, we constructed an optimized procedure for identifying highly cytotoxic antibodies from a variety of the T-cell–recruiting antibodies engineered from a series of antibodies against cancer antigens of epidermal growth factor receptor family and T-cell receptors. By developing and applying a set of rapid operations for expression vector construction and protein preparation, we screened the cytotoxicity of 104 small antibodies with diabody format and identified some with 10³-times higher cytotoxicity than that of previously reported active diabody. The results demonstrate that cytotoxicity is enhanced by synergistic effects between the target, epitope, binding affinity, and the order of heavy-chain and light-chain variable domains. We demonstrate the importance of screening to determine the critical rules for highly cytotoxic antibodies.