Identification of convergent antibody specificity sequence patterns

A method, comprising:

• providing to a candidate identification system a plurality of input amino acid sequences from a library of immunoglobulin heavy chain variable region (VH) sequences or a library of immunoglobulin light chain variable (VL) region sequences,
• wherein the library comprises an entire immunoglobulin repertoire from animals immunized with an antigen;
• transforming, by an encoder executed by the candidate identification system, the plurality of input amino acid sequences into a latent space,
• wherein the encoder uses a linear classifier, a plurality of rectified linear units, and a variational autoencoder (VAE) to transform the plurality of input amino acid sequences into the latent space, and
• wherein the linear classifier is a linear support vector machine (SVM) classifier of the immunized antigen exposure;
• determining, by a clustering engine executed by the candidate identification system, a plurality of sequence clusters within the latent space;
• identifying, by the clustering engine, a convergent sequence cluster that is predictive for the immunized antigen exposure;
• selecting, by a candidate generation engine executed by the candidate identification system, within the latent space an immunized antigen-associated convergent sequence cluster;
• generating, by the candidate generation engine using a decoder, a candidate amino acid sequence based on the immunized antigen-associated convergent sequence cluster,
• wherein the decoder comprises a plurality of different neural networks comprising a plurality of long short-term recurrent neural networks, and
• wherein generating the candidate amino acid sequence further comprises providing the immunized antigen-associated convergent sequence cluster to each of the plurality of long short-term recurrent neural networks;
• and
• generating a library of hybridoma cell clones expressing antibodies comprising the candidate amino acid sequence.