Inter-sample comparisons of the T cell receptor (TCR) repertoire are crucial for gaining a better understanding into the immunological states determined by different collections of T cells from different donor sites, cell types, and genetic and pathological backgrounds. As a theoretical approach for the quantitative comparison, previous studies utilized the Poisson abundance models and the conventional methods in ecology, which focus on the abundance distribution of observed TCR sequences. However, these methods ignore the details of the measured sequences and are consequently unable to identify sub-repertoires that might have the contributions to the observed inter-sample differences. In this paper, we propose a new comparative approach based on TCR sequence information, which can estimate the low-dimensional structure by projecting the pairwise sequence dissimilarities in high-dimensional sequence space. The inter-sample differences are then quantified according to information-theoretic measures among the distributions of data estimated in the embedded space. Using an actual dataset of TCR sequences in transgenic mice that have strong restrictions on somatic recombination, we demonstrate that our proposed method can accurately identify the inter-sample hierarchical structure, which is consistent with that estimated by previous methods based on abundance or count information. Moreover, we identified the key sequences that contribute to the pairwise sample differences. Such identification of the sequences contributing to variation in immune cell repertoires may provide substantial insight for the development of new immunotherapies and vaccines.