Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
HeMeNet: Heterogeneous Multichannel Equivariant Network for Protein Multi-task Learning
Authors: Rong Han, Wenbing Huang, Lingxiao Luo, Xinyan Han, Jiaming Shen, Zhiqiang Zhang, Jun Zhou, Ting Chen
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive evaluations on our benchmark verify the effectiveness of multi-task learning, and our model surpasses state-of-the-art models. Experiments In this section, we will first introduce the experimental setup. Next, we evaluate our model on the proposed dataset Protein-MT for affinity and property prediction in both single-task and multi-task settings and compare it with other baseline models. Then, we experiment with different readout strategies and compare their performance. At last, we perform ablation experiments on different modules. |
| Researcher Affiliation | Collaboration | 1 BNRist, Department of Computer Science and Technology, Tsinghua University 2 Gaoling School of Artificial Intelligence, Renmin University of China 3 Ant Group CO., Ltd. |
| Pseudocode | No | The paper describes the methodology using mathematical equations and descriptive text, for example in 'Heterogeneous Multi-channel Equivariant Message Passing' and 'Task-Aware Readout' sections, but does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code https://github.com/hanrthu/He Me Net |
| Open Datasets | Yes | We carefully integrate the structures and labels from 4 public datasets with our proposed standard process and construct a new benchmark named Protein Multiple Tasks (Protein-MT), which consists of 6 representative tasks upon 3 different types of inputs. The LBA and PPA tasks originated from the PDBbind database (Wang et al. 2004). The EC task is constructed by (Gligorijevi c et al. 2021b). The GO task aims to predict the hierarchically related functional properties of gene products (Gligorijevi c et al. 2021b): Molecular Function (MF), Biological Process (BP), and Cellular Component (CC). |
| Dataset Splits | Yes | After our above matching process, we formulate the train/validation/test split in terms of the chain-level sequence identity through the alignment methods commonly used in single-chain property prediction tasks (Gligorijevi c et al. 2021b). |
| Hardware Specification | No | The paper discusses the experimental setup, baselines, and evaluation metrics, but it does not provide any specific details about the hardware (e.g., GPU model, CPU model, memory) used for running the experiments. |
| Software Dependencies | No | The paper discusses the experimental setup, baselines, and evaluation metrics, but it does not specify any software or library names with version numbers that would be needed to replicate the experiment. |
| Experiment Setup | No | The 'Experimental Setup' section describes the task settings (single-task vs. multi-task), baselines, and evaluation metrics. However, it does not provide specific experimental setup details such as concrete hyperparameter values (e.g., learning rate, batch size, number of epochs) or optimizer settings. |