AbODE: Ab initio antibody design using conjoined ODEs
Authors: Yogesh Verma, Markus Heinonen, Vikas Garg
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The proposed model significantly outperforms existing methods on standard metrics across benchmarks.4. Experiments Tasks We benchmark Ab ODE on a series of challenging tasks: (i) we evaluate the model on unconditional antibody sequence and structure generation against ground truth structures in the Structural Antibody Database SAb Dab (Dunbar et al., 2014) section 4.1, (ii) we benchmark our method in terms of its ability to generate antigen-conditioned antibody sequences and structures from SAb Dab in section 4.2, (iii) we evaluate our model on the task of designing CDRH3 over 60 manually selected diverse complexes (Adolf Bryfogle et al., 2018) in section 4.3, (iv) we extend our model to incorporate information about the constant region of the antibody in section 4.4, and finally, (v) we extend Ab ODE to de novo protein sequence design with a fixed backbone in section 4.5. |
| Researcher Affiliation | Collaboration | 1Department of Computer Science, Aalto University, Finland 2Yai Yai Ltd. Correspondence to: Yogesh Verma <yogesh.verma@aalto.fi>. |
| Pseudocode | No | The paper describes the method using mathematical formulations and descriptive text, but does not include explicit pseudocode or an algorithm block. |
| Open Source Code | No | The paper does not provide any explicit statement about open-sourcing code or a link to a code repository for the described methodology. |
| Open Datasets | Yes | We obtained the antibody sequences and structure from Structural Antibody Database (SAb Dab) (Dunbar et al., 2014)We use the CATH 4.2 dataset curated by Ingraham et al. (2019b) |
| Dataset Splits | Yes | We then randomly split the clusters into training, validation, and test sets with an 8:1:1 ratio. |
| Hardware Specification | No | The calculations were performed using resources made available by the Aalto University Science-IT project. |
| Software Dependencies | No | Ab ODE is implemented in Py Torch (Paszke et al., 2019). |
| Experiment Setup | Yes | Implementation Ab ODE is implemented in Py Torch (Paszke et al., 2019). We used three layers of a Transformer Convolutional Network (Shi et al., 2020) with embedding dimensions of 128 256 64. Our models were trained with the Adam optimizer for 5000 epochs using batch size 300. For details, we refer the reader to Appendix A.3.We used three layers of Transformer Convolutional Network with hidden embedding dimensions of 128 256 64. The ODE solver operated over time-steps t [0, 200], where we took the last time step value as the final prediction of the model. The ODE system is solved with the Adaptive heun solver with an adaptive step size. We train the models for 10000 epochs with the Adam optimizer and use a batch size of 300. |