Journal: bioRxiv

Article Title: Protein Deimmunization with a Generative Diffusion Model

doi: 10.64898/2026.10.09.758136

Figure Lengend Snippet: (A) Fixed amino acid residues for MESSI-based deimmunization of LuxSit-i. The structure of LuxSit-i (shown in blue and red) with its luciferin substrate (shown in green) is generated by Alphafold-3. The color on LuxSit-i denotes the number of overlapping NMP-predicted HLA-DRB1*15:01 epitopes at each amino acid position. Fixed amino acid residues are depicted as yellow-colored spheres. (B) Sequence identity of MESSI-generated LuxSit-i variants relative to the parental LuxSit-i. (C) Structural metrics of MESSI-generated LuxSit-i variants. (D) Comparison of epitope density between parental LuxSit-i and a MESSI-generated variant, MESSI-D2. All structures were generated by AlphaFold-3. Color denotes the number of overlapping NMP-predicted HLA-DRB1*15:01 epitopes at each amino acid position. (E) Experimental validation of luminescence from MESSI-generated LuxSit-i variants. Gene fragments encoding the parental and MESSI-designed proteins were synthesized and used as linear DNA template for in vitro RNA transcription and protein translation (IVTT). Afterwards, the reaction product was purified with a nickel spin column to remove the His-tagged IVTT machinery. A final concentration of 25 µM of diphenylterazine (DTZ) was added to the purified proteins. Luminescence was recorded on a plate reader every minute over 2 hr. (F) Peak luminescence recorded within 2 hours after DTZ addition. “No template” denotes the negative control condition where no DNA template was added to the IVTT reaction. Error bars, standard deviation. n = 3 replicates. A two sided unpaired t-test with welch’s correction is calculated between no-target condition and each of the MESSI designs and parent. **, p < 0.01; ****, p < 0.0001. (G) Computational workflow for designing a deimmunized binder against PD-1 using MESSI coupled with BindCraft. In particular, MESSI is inserted at the end of the original BindCraft workflow to replace the soluble MPNN inverse folding model to design sequences against the optimized structures. (H) Structural metrics of MESSI-generated variants of anti-PD-1 binder. (I) Comparison of epitope density between the parental anti-PD-1 binder and a MESSI-generated variant, MESSI-D4. All structures were generated by AlphaFold-3. Blue colors denote a binder region that does not contain NMP predicted epitopes. Red color denotes binder regions with NMP-predicted HLA-DRB1*15:01 epitopes greater than 0. White color denotes the PD-1, the target protein for the designed binder. (J) Strep-tagged minibinders are produced with Genscript IVTT kit. After the production, the binders are purified with streptavidin beads and eluted in 1 mL of PBS. The dot plots show the concentration (expression) fold-change (FC) of the MESSI-generated binders relative to the parental binder designed by BindCraft. (K) Bio-Layer Interferometry (BLI) binding curves for measuring the binding parameters of MESSI-designed anti-PD-1 binders against the target PD-1 protein. The assay was performed at 30 °C and at 1,000 rpm. His-tagged PD-1 protein was firstly immobilized onto HIS1K biosensor on Octet RED384. Binders was applied at a concentration of 1 µM as an analyte for association and dissociation steps for 150 seconds and 30 seconds respectively. PA denotes the parent binder. (L) Equilibrium dissociation constant (K d ) of anti-PD-1 binder variants. K d was derived from the association (K a ) and dissociation (K dis ) rate constants measured from the five concentration scan BLI curves. Error bars represent the 95% confidence interval of the fitted K d . Differences in K d were assessed by a two-sided z-test using moving-block bootstrap standard errors of the global fit (85 seconds blocks, 1,000 resamples; Holm-corrected for comparisons). **, p<0.01

Article Snippet: Binding kinetics were measured by biolayer interferometry (BLI) on an Octet RED384 instrument (Sartorius, Cat. No. FB-20493) using anti-penta-HIS (HIS1K) biosensors (Sartorius, Cat. No. 18-5120) and tilted-bottom 384-well microplates (Sartorius, Cat. No. 18-5166).

Techniques: Generated, Sequencing, Comparison, Variant Assay, Biomarker Discovery, Synthesized, In Vitro, Purification, Concentration Assay, Negative Control, Standard Deviation, Produced, Expressing, Binding Assay, Affinity Constant, Derivative Assay, Blocking Assay