Skip to content

Commit 93b4809

Browse files
authored
Merge pull request #6 from huangtinglin/master
Add faformer project page
2 parents 6a80a68 + eb557cf commit 93b4809

File tree

8 files changed

+76
-5
lines changed

8 files changed

+76
-5
lines changed
489 KB
Loading

app/projects/faformer/fig/exp1.png

78.7 KB
Loading

app/projects/faformer/fig/exp2.png

116 KB
Loading

app/projects/faformer/fig/fa.png

211 KB
Loading
295 KB
Loading
277 KB
Loading

app/projects/faformer/page.mdx

Lines changed: 71 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,71 @@
1+
import { Authors, Badges } from '@/components/utils'
2+
3+
# Protein-Nucleic Acid Complex Modeling with Frame Averaging Transformer
4+
5+
<Authors
6+
authors="Tinglin Huang, Yale University; Zhenqiao Song, Carnegie Mellon University; Rex Ying, Yale University; Wengong Jin, Northeastern University, Broad Institute of MIT and Harvard"
7+
/>
8+
9+
<Badges
10+
venue="NeurIPS 2024"
11+
github="https://github.com/Graph-and-Geometric-Learning/Frame-Averaging-Transformer"
12+
arxiv="https://arxiv.org/abs/2406.09586"
13+
pdf="https://arxiv.org/pdf/2406.09586"
14+
/>
15+
16+
17+
## Introduction
18+
Understanding and predicting how protein forms a complex with nucleic acid/protein offers insights into binding affinity, which can be applied to high-throughput screening of drug candidates, such as aptamers (RNA-based drugs), **without** labels, as shown in Figure 1.
19+
20+
![Figure 1: Comparison between two aptamer candidates on GFP.|scale=0.5](./fig/binding_comparison.png)
21+
22+
Motivated by this, we propose **contact map prediction-based aptamer screening paradigm**. Specifically, as presented in Figure 2(a), our model is trained to identify specific contact pairs between residues and nucleotides when forming a complex. The maximum contact probability across all pairs is then interpreted as the binding affinity, which is subsequently used for aptamer screening.
23+
24+
![Figure 2: (a) The pipeline of contact map prediction between protein and nucleic acid, and applying the predicted results for screening in an unsupervised manner. (b) Comparison between Transformer with vanilla frame averaging framework and FAFormer, where the blue cells indicate FA-related modules. |scale=0.5](./fig/overview.png)
25+
26+
Learning E(3) equivariant transformation is the key factor to modeling the protein/nucleic acid 3D structures. In this paper, we propose **FAFormer**, an equivariant Transformer architecture that integrates FA as a geometric module within each layer, as shown in Figure 2(b). FA as a geometric component offers flexibility to effectively integrate geometric information into node representations while preserving the spatial semantics of coordinates and without major modicification on the architectures. FAFormer opens new possibilities for designing equivariant architectures in this domain.
27+
28+
29+
## Method
30+
31+
### Frame Averaging (FA)
32+
33+
Frame averaging (FA) is an encoder-agnostic framework that can make a given encoder equivariant to the Euclidean symmetry group. Specifically, FA proposes to model
34+
the coordinates in eight different frames extracted by PCA, achieving equivariance by averaging the encoded representations, as presented in Figure 3.
35+
36+
![Figure 3: Frame Averaging.|scale=0.5](./fig/fa.png)
37+
38+
You can consider FA as a model "wrapper", where the model architecture doesn't need to be modified but would seperately process 8 inputs. We use $f_{\mathcal{F}}(\mathbf{X})=\{\mathbf{X}^{(g)}\}_{\mathcal{F}}$ to denote the FA operation, where $\mathbf{X}^{(g)}$ is the input in the $g$-th frame. Besides, we use $f_{\mathcal{F}^{-1}}(\{\mathbf{\hat{X}}^{(g)}\}_{\mathcal{F}})=\hat{X}$ to represent the inverse mapping, which is an E(3) equivarnat operation. Note that $\hat{X}^{(g)}$ could be obtained from the encoder. The outcome could be invariant when simply averaging the representations without inverse matrix.
39+
40+
Building on such foundation, we generalize the vanilla Linear module to FA Linear module:
41+
$$
42+
\text{Linear}_{\mathcal{F}}(\mathbf{X})=\frac{1}{8}\sum_{g}\mathbf{X}^{(g)}\mathbf{W}_g
43+
$$
44+
where $\mathbf{W}_g\in\Bbb{R}^{3\times D}$. Note that the output of FA Linear module is E(3) invariant.
45+
46+
47+
### Overall architecture of FAFormer
48+
49+
![Figure 4: Overview of FAFormer architecture. The input consists of the node features, coordinates, and edge representations.|scale=0.5](./fig/faformer.png)
50+
51+
As shown in Figure 4(a), the input of FAFormer comprises the node features $\mathbf{Z}\in\Bbb{R}^{N\times D}$, coordinates $\mathbf{X}\in\Bbb{R}^{N\times 3}$, and edge representations $\mathbf{E}\in\Bbb{R}^{N\times K\times D}$ where $K$ is the number of nearest neighbors. Each core modules are dedicatedly integrated with FA, including
52+
53+
* Local frame edge module focuses on local spatial context by constructing the frames on the point cloud centered around each node;
54+
* As shown in Figure 4(f), the edge module builds frame locally around each node’s neighbors, which is compatible to the self-attention operation.
55+
* Biased MLP attention module applies FA to enable equivariant multi-head attention on the geometric features;
56+
* The multi-head attention on coordinate is conduct based on FA.
57+
* Global frame FFN extends the FFN by incorporating geometric information within node representations using FA.
58+
59+
## Experiments
60+
### Contact Map Prediction
61+
62+
This task aims to predict the exact contact pairs between protein and protein/nucleic acids, which conducts binary classification over all pairs. This task is challenge due to the sparsity of the contact pairs. We compare FAFormer with six state-of-the-art models, and the results are presented in Table 1.
63+
64+
![Figure 5: Contact Map Prediction.|scale=0.5](./fig/exp1.png)
65+
66+
### Unsupervised Aptamer Screening
67+
68+
This task aims to screen the positive aptamers from a large number of candidates for a given protein target. We quantify the binding affinities between RNA and the protein target as the highest contact probability among the residue-nucleotide pairs. The models are first trained on the protein-RNA complexes training set using the contact map prediction, then the aptamer candidates are ranked based on the calculated highest contact probabilities. Top-10 precision, Top-50 precision, and PRAUC are used as the metrics.
69+
70+
![Figure 6: Aptamer screening.|scale=0.5](./fig/exp2.png)
71+

config/publications.ts

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -19,13 +19,13 @@ export const publications = [
1919
{
2020
title: "Protein-Nucleic Acid Complex Modeling with Frame Averaging Transformer",
2121
authors: "Tinglin Huang, Zhenqiao Song, Rex Ying, Wengong Jin",
22-
venue: "MLSB Workshop, NeurIPS, 2023",
23-
page: null,
22+
venue: "NeurIPS, 2024",
23+
page: "faformer",
2424
code: "https://github.com/Graph-and-Geometric-Learning/Frame-Averaging-Transformer",
2525
paper: "https://arxiv.org/abs/2406.09586",
2626
abstract: "Propose a novel unsupervised aptamer screening paradigm and FAFormer, a frame averaging-based equivariant transformer architecture.",
2727
impact: "We explore a new angle to conduct aptamer screening in an unsupervised manner by leveraging the strong correlation with the contact map prediction task. Besides, we propose to integrate Frame Averaging (FA) within each transformer module and develop FAFormer, highlighting a new possibility for geometric encoder design in this domain.",
28-
tags: [Tag.Applications],
28+
tags: [Tag.Applications, Tag.GeometricAndGraphLearning],
2929
},
3030
{
3131
title: "From Similarity to Superiority: Channel Clustering for Time Series Forecasting",
@@ -54,8 +54,8 @@ export const publications = [
5454
},
5555
{
5656
title: "HEART: Learning Better Representation of EHR Data with a Heterogeneous Relation-Aware Transformer",
57-
authors: "T Huang, SA Rizvi, R Krishna Thakur, V Socrates, M Gupta, D Dijk, RA Taylor, R Ying",
58-
venue: "Preprint",
57+
authors: "Tinglin Huang, Syed Asad Rizvi, Rohan Krishna Thakur, Vimig Socrates, Meili Gupta, David van Dijk, R. Andrew Taylor, Rex Ying",
58+
venue: "Journal of Biomedical Informatics 159 (2024): 104741",
5959
page: null,
6060
code: null,
6161
paper: "https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4904741",

0 commit comments

Comments
 (0)