1: \begin{abstract}

2:        The insertion, deletion, substitution (IDS) correcting code has garnered increased attention 

3:        due to significant advancements in DNA storage that emerged recently. 

4:        Despite this, the pursuit of optimal solutions in IDS-correcting codes remains an open challenge, 

5:        drawing interest from both theoretical and engineering perspectives.

6:        This work introduces a pioneering approach named THEA-code. 

7:        The proposed method follows a heuristic idea of employing an end-to-end autoencoder for 

8:        the integrated encoding and decoding processes. 

9:        To address the challenges associated with deploying an autoencoder as an IDS-correcting code, 

10:        we propose innovative techniques, including the differentiable IDS channel, the entropy constraint on the codeword, and 

11:        the auxiliary reconstruction of the source sequence. 

12:        These strategies contribute to the successful convergence of the autoencoder, 

13:        resulting in a deep learning-based IDS-correcting code with commendable performance.

14:        Notably, THEA-Code represents the first instance of a deep learning-based code 

15:        that is independent of conventional coding frameworks in the IDS-correcting domain. 

16:        Comprehensive experiments, including an ablation study, provide a detailed analysis and affirm the effectiveness of THEA-Code. 

17: \end{abstract}

18: