Deep generative models for T cell receptor protein sequences

Abstract

Probabilistic models of adaptive immune repertoire sequence distributions can be used to infer the expansion of immune cells in response to stimulus, differentiate genetic from environmental factors that determine repertoire sharing, and evaluate the suitability of various target immune sequences for stimulation via vaccination. Classically, these models are defined in terms of a probabilistic V(D)J recombination model which is sometimes combined with a selection model. In this paper we take a different approach, fitting variational autoencoder (VAE) models parameterized by deep neural networks to T cell receptor (TCR) repertoires. We show that simple VAE models can perform accurate cohort frequency estimation, learn the rules of VDJ recombination, and generalize well to unseen sequences. Further, we demonstrate that VAE-like models can distinguish between real sequences and sequences generated according to a recombination-selection model, and that many characteristics of VAE-generated sequences are similar to those of real sequences.

On generative models of T-cell receptor sequences

“…We compare their performances for predicting the distribution of TCR sequences in controlled conditions, training and validating on the same datasets. Contrary to the claims of the original VAE paper [21], we show that that knowledge guided models perform as well as the variational auto-encoder or even better, at a lower computational cost.…”

“…In addition, because no software implementation of the selection model was provided with the original article [19], Davidsen et al [21] compared their VAE approach to a reduced version of this selection model (not examined in [19]), which they call OLGA.Q. In that model, only VJ usage and CDR3 length were included: h i,L = 0.…”

“…In Ref. [21], a variant of the traditional auto-encoder detailed in [24] was used. Here we focus on the version of the VAE called basic in that paper.…”

“…Recently, Davidsen et al [21] described an elegant approach for learning the distribution of T-cell receptor beta sequences (TCRβ or simply TCR in the follow-ing), based on a Variational Auto-Encoder (VAE). The method makes it possible to generate new sequences with the same statistics as real repertoires, and to evaluate the frequency of individual sequences, which agree with the data with good accuracy.…”

“…Here we compare the VAE method [21] with the previously proposed model of generation and selection, called SONIA [19,20]. We compare their performances for predicting the distribution of TCR sequences in controlled conditions, training and validating on the same datasets.…”