---
title: 'TCPNet: Topology-Complete Perceptron Network'
url: https://www.emergentmind.com/topics/topology-complete-perceptron-network-tcpnet
type: topic
---

# TCPNet: Topology-Complete Perceptron Network

Searching arXiv for the primary TCPNet paper and closely related topological neural network work.
The **Topology-Complete Perceptron Network (TCPNet)** is an **SE(3)-equivariant** topological neural network introduced within the **Topotein** framework for protein representation learning. In that formulation, TCPNet operates on a **Protein Combinatorial Complex (PCC)** whose ranks encode residues, residue-interaction edges, secondary structure elements (SSEs), and the whole protein, and it uses the **Topology-Complete Perceptron (TCP)** module as its basic building block [2509.03885]. The designation “topology-complete” does not appear in that work as a formal universality theorem; rather, it denotes an architectural commitment to exploiting the **full topological structure of the PCC across all ranks**, instead of restricting computation to a residue graph alone [2509.03885].

## 1. Definition, scope, and nomenclature

In the Topotein formulation, TCPNet is the main encoder built on the PCC. It is defined as a network that is **SE(3)-equivariant**, operates on **multi-rank cells**—residues, residue-interaction edges, secondary structure elements, and whole proteins—and performs a **four-step hierarchical message passing scheme** across ranks \(0\)–\(3\) [2509.03885]. Formally, the input is a PCC
\[
\mathcal{C}=(S,\mathcal{X},\mathrm{rk}),
\]
with scalar and vector features at each rank \(r\in\{0,1,2,3\}\), and the output is a protein-level embedding, optionally accompanied by residue-level embeddings for downstream tasks [2509.03885].

The motivation stated for TCPNet is specific to protein representation learning. The authors identify three deficiencies in prior model classes: sequence-only protein language models do not use 3D structure directly and do not explicitly model SSEs or multi-scale geometric hierarchy; standard residue-level geometric graph neural networks (GGNNs) create information bottlenecks between SSEs and do not provide a unified representation of residues, interactions, SSEs, and proteins; and existing topological deep learning models based on simplicial, cellular, or hypergraph constructions either impose strict boundary constraints or do not encode hierarchy in a protein-specific way [2509.03885].

Within the broader literature, TCPNet is best understood as a domain-specific instance of a topological neural network rather than a generic synonym for all topology-aware perceptron systems. The general idea of a neural network whose inputs are drawn from a topological space rather than a finite-dimensional Euclidean space was formalized earlier as a **Topological Neural Network (TNN)**, with related **Distributional Neural Networks (DNNs)** on spaces of measures [2305.16639]. That earlier work is close in spirit, but it addresses universal approximation on Tychonoff spaces and spaces of measures, whereas TCPNet in Topotein is a concrete protein architecture on combinatorial complexes [2305.16639].

## 2. Protein Combinatorial Complex as the computational domain

The PCC is a **combinatorial complex**
\[
(S,\mathcal{X},\mathrm{rk}),
\]
where \(S\) is a finite vertex set, \(\mathcal{X}\subseteq\mathcal{P}(S)\setminus\{\emptyset\}\) is a set of cells, and \(\mathrm{rk}:\mathcal{X}\to\mathbb{Z}_{\ge 0}\) is an order-preserving rank function such that
\[
x\subseteq y \Rightarrow \mathrm{rk}(x)\le\mathrm{rk}(y),
\]
with every singleton \(\{s\}\) assigned rank \(0\) [2509.03885]. Unlike simplicial complexes, the PCC imposes **no requirement that the boundary of a cell be present**, which is presented as crucial for biological structures such as irregular or non-contiguous SSEs [2509.03885].

For proteins, the rank set is \(R=\{0,1,2,3\}\). The specialization is as follows [2509.03885]:

| Rank | Cell type | Interpretation |
|---|---|---|
| \(0\) | residues | individual residues |
| \(1\) | directed edges | residue interactions |
| \(2\) | SSEs | secondary structure elements |
| \(3\) | protein cell | the whole protein |

The PCC uses **directed pairwise edges** as 1-cells rather than undirected hyperedges. Each residue is connected to its **16 nearest neighbors** in 3D, and directionality permits asymmetric incidence between residues and edges [2509.03885]. Rank-2 cells are SSEs formed from **sequentially consecutive residues of the same DSSP label** with minimum length \(3\), and each residue belongs to at most one SSE, so the 2-cells do not overlap [2509.03885]. Rank \(3\) is a single protein cell containing all residues [2509.03885].

Incidence matrices organize the relations between ranks. For \(r\neq r'\),
\[
\mathcal{B}^{r\to r'} \in \{0,1\}^{|N_r|\times |N_{r'}|},
\]
and the corresponding adjacency construction is
\[
\mathcal{L}^{r\sim r'} = \mathcal{B}^{r\to r'}\cdot \mathcal{B}^{r'\to r},\qquad
\mathcal{A}^{r\sim r'} = \mathcal{L}^{r\sim r'} - \mathcal{D}^{r\sim r'}.
\]
These matrices support within-rank and cross-rank neighborhood definitions for message passing [2509.03885].

A distinctive PCC construction is the **outer-edge neighborhood**, introduced because non-overlapping SSEs cannot communicate by shared residues. It is defined by
\[
\mathcal{N}^{2 \rightarrow 1}_{outer} = \mathcal{B}^{2 \rightarrow 0} \cdot \mathcal{B}^{0 \rightarrow 1} - \mathcal{B}^{2 \rightarrow 1},
\tag{1}
\]
\[
(\mathcal{N}^{1 \rightarrow 2}_{outer})^\top = \mathcal{B}^{2 \rightarrow 0} \cdot (\mathcal{B}^{1 \rightarrow 0})^\top - \mathcal{B}^{2 \rightarrow 1}.
\tag{2}
\]
These matrices identify edges that originate in an SSE and exit it, or enter it from outside, while excluding internal SSE edges [2509.03885]. In practice, TCPNet uses the intersection
\[
\mathcal{N}^{2 \to 1} = \mathcal{N}^{2 \rightarrow 1}_{outer} \cap (\mathcal{N}^{1 \rightarrow 2}_{outer})^\top
\]
for scalarization at the SSE level [2509.03885].

The PCC is also hierarchical at the feature level. Residues carry amino-acid, angular, positional, and geometric vector features; edges carry distance and displacement features; SSEs carry type, size, angular, eigenspectral, and shape-descriptor features; and the protein cell carries global statistics such as size, covariance eigenvalues, shape descriptors, radius of gyration, and contact density [2509.03885]. This hierarchy is explicitly combinatorial and geometric.

## 3. The Topology-Complete Perceptron module and SE(3)-equivariant processing

The **Topology-Complete Perceptron (TCP)** is the fundamental layer used by TCPNet. It is described as an **SE(3)-equivariant adaptation of the GCP module from GCPNet**, generalized to arbitrary PCC ranks [2509.03885]. For a cell of rank \(r\), the TCP takes scalar features \(\mathbf{h}_s\), vector features \(\mathbf{h}_v\in\mathbb{R}^{d_v\times 3}\), and a rank-specific local frame \(\mathcal{F}^{(r)}_i\), or in practice associated edge frames [2509.03885].

The module performs vector-channel processing and scalarization through three stages [2509.03885]:
\[
s = \sigma(V_s(\mathbf{h}_v)) \in \mathbb{R}^{3\times 3},
\]
\[
z = \sigma(V_d(\mathbf{h}_v)) \in \mathbb{R}^{\frac{d_v}{\lambda}\times 3},
\]
followed by
\[
\mathbf{h}_s' = (\mathbf{h}_s, \mathcal{S}^{(r)}_i(s), \|z\|_2),
\]
and output maps
\[
\mathbf{h}_{s,out} = \sigma(S_{out}(\mathbf{h}_s')),
\]
\[
\mathbf{h}_v' = \sigma(V_u(z)),
\]
\[
\mathbf{h}_{v,out} = \mathbf{h}_v' \odot \sigma_g(S_{gate}(\mathbf{h}_{s,out})).
\]
The scalar gate modulates vector channels while preserving equivariance, because vector features remain geometric vectors and only scalar functions are used to scale them [2509.03885].

A central implementation choice is **edge-centric scalarization**. For an edge \((i,j)\), the local frame is
\[
\mathcal{F}^{(1)}_{(i,j)} =
\left(
\hat{\mathbf{e}_{ij}},
\widehat{\mathbf{x}_j\times\mathbf{x}_i},
\hat{\mathbf{e}_{ij}} \times \widehat{\mathbf{x}_j\times\mathbf{x}_i}
\right),
\]
with \(\mathbf{e}_{ij}=\mathbf{x}_j-\mathbf{x}_i\), and scalarization is
\[
\mathcal{S}^{(1)}_{(i,j)}(\mathbf{h}^{(1)}_{(i,j),v})
= \operatorname{flatten}\left(\mathbf{h}^{(1)}_{(i,j),v} \cdot \mathcal{F}^{(1)}_{(i,j)}\right).
\]
For nodes and SSEs, scalarization averages projections onto incident edge frames or outer-edge frames, respectively [2509.03885]. At rank \(3\), the protein frame is obtained from PCA eigenvectors of residue positions, with sign disambiguation using the farthest residue direction [2509.03885].

The paper attributes **SE(3)-equivariance** to three ingredients: vector features are treated as geometric vectors, scalarized quantities are obtained relative to local orthonormal frames and are therefore rotation- and translation-invariant, and vector computations remain linear with scalar-only gating [2509.03885]. A plausible implication is that equivariance is not attached to any single rank in isolation, but is maintained across the entire rank hierarchy through a common scalar–vector processing template.

## 4. Four-step hierarchical message passing in TCPNet

TCPNet consists of an embedding module, repeated interaction layers, residual normalization, and a readout stage [2509.03885]. The interaction stack has \(L=6\) layers in the reported implementation, and each interaction layer executes a **four-step hierarchical message passing scheme** [2509.03885].

**Step 1: edge-level messaging.** For each directed edge \((i,j)\), the network computes a message from the source residue, target residue, edge features, and the SSE features of the parent SSEs of the two residues:
\[
\mathbf{m}_{ij} = \phi^{(1)}\bigl(
\mathbf{h}^{(0)}_i, \mathbf{h}^{(0)}_j,
\mathbf{h}^{(1)}_{ij}, \mathbf{n}_i, \mathbf{n}_j
\bigr),
\]
where \(\mathbf{n}_i\) is the parent-SSE feature if residue \(i\) belongs to an SSE and \(\mathbf{0}\) otherwise [2509.03885]. The edge state itself is updated by a TCP block with residual structure,
\[
\mathbf{h}^{l+1}_{ij} = \varphi_{\text{msg}^l}(\mathbf{h}^l_{ij}) + \mathbf{h}^l_{ij},
\]
and then transformed into a final message with scalar attention on the scalar channel [2509.03885].

**Step 2: SSE-level aggregation.** Each SSE aggregates from four sources: its own features, constituent residues through \(\mathcal{B}^{2\to 0}\), internal edges through \(\mathcal{B}^{2\to 1}\), and outer-edge messages through \(\mathcal{N}^{2\to 1}\):
\[
\mathbf{u}^{(2)}_{i} = \phi^{(2)}\left(
\mathbf{h}^{(2)}_i,
\bigoplus_{j\in \mathcal{B}^{2\to 0}_i} \mathbf{h}^{(0)}_j,
\bigoplus_{j\in \mathcal{B}^{2\to 1}_i} \mathbf{h}^{(1)}_{j},
\bigoplus_{(j,k)\in \mathcal{N}^{2\to 1}_i} \mathbf{m}_{jk}
\right).
\]
This is the stage at which inter-SSE contact geometry is explicitly incorporated without collapsing SSEs into coarse supernodes [2509.03885].

**Step 3: residue refinement.** Updated SSE information is pushed back to residues, and incoming edge messages are aggregated:
\[
\mathbf{u}^{(0)}_i = \phi^{(0)}\left(
\mathbf{h}^{(0)}_i,
\mathbf{m}_i^{(2)},
\bigoplus_{j\in (\mathcal{B}^{1\to 0}_i)^\top}\mathbf{m}_{ji}
\right).
\]
This stage fuses local residue context with the meso-scale structural context represented at the SSE level [2509.03885].

**Step 4: protein-level update.** The protein cell is updated from the lower-rank states,
\[
\mathbf{u}^{(3)}_i = \phi^{(3)}\bigl(
\mathbf{u}^{(0)}_i,
\mathbf{u}^{(2)}_i,
\mathbf{h}^{(3)}_i
\bigr),
\]
with aggregation over residues and SSEs in practice [2509.03885].

Residual updates with GVP-style layer normalization are then applied rank-wise:
\[
\mathbf{h}^{(r)'}_i = \text{LN}(\mathbf{u}^{(r)}_i + \mathbf{h}^{(r)}_i),\qquad r\in\{0,2,3\},
\]
while edges are updated through their own TCP-plus-residual pathway [2509.03885]. Readout can be performed either by pooling residue embeddings or by using the rank-\(3\) protein embedding directly [2509.03885].

At the level of topological deep learning formalism, TCPNet instantiates a generic neighborhood-based update
\[
\mathbf{h}_i^{l+1} = \phi\left(
  \mathbf{h}_i^l,\;
  \bigotimes_{\mathcal{N}\in\mathfrak{N}}
    \bigoplus_{j\in\mathcal{N}_i}
       \psi_{\mathcal{N}}(\mathbf{h}_i^l,\mathbf{h}_j^l) \right),
\]
where \(\mathfrak{N}\) contains incidence, adjacency, and outer-edge neighborhood types on the PCC [2509.03885]. This places TCPNet within the general message-passing paradigm for topological neural networks on combinatorial complexes.

## 5. Experimental performance and task profile

TCPNet was evaluated on four protein representation learning tasks: **inverse folding** on CATH 4.4, **fold classification** on SCOP 1.75, **cellular component prediction** on Gene Ontology data, and **antibody developability** on SabDab [2509.03885]. The reported training configuration used **Adam** with learning rate \(0.001\), **ReduceLROnPlateau** with factor \(0.6\) and patience \(5\), early stopping with patience \(10\), batch size \(32\), and NVIDIA A100 80GB hardware [2509.03885].

Against **GCPNet**, **GVP-GNN**, and **EGNN**, TCPNet showed a task-dependent profile [2509.03885]. On inverse folding, GVP-GNN remained best with perplexity \(5.280\) and accuracy \(0.474\), while TCPNet was second-best with perplexity \(5.822\) and accuracy \(0.441\) [2509.03885]. On fold classification, TCPNet was best on all three SCOP test splits:
- Fold: \(0.433\)
- Superfamilies: \(0.558\)
- Families: \(0.971\)

The paper notes that on the hardest **Fold** split, TCPNet improves over second-best EGNN from \(0.401\) to \(0.433\), approximately \(3\%\) absolute [2509.03885]. On **cellular component prediction**, TCPNet was best in the **structure-only** setting with \(F1_{\max}=0.392\), though slightly behind GCPNet and GVP-GNN when sequence features were added [2509.03885]. On **antibody developability**, TCPNet was best in the structure-only case with **AUPRC** \(0.854\), and second-best with structure plus sequence at \(0.874\), narrowly behind GCPNet at \(0.878\) [2509.03885].

A second comparison examined topological enhancements relative to their GGNN baselines. **TCPNet** consistently improved over **GCPNet** across all reported tasks, including inverse folding, fold classification, cellular component prediction, and antibody developability [2509.03885]. By contrast, **GVP-TNN** improved strongly on fold classification but degraded on inverse folding and some other tasks, while **ETNN** was significantly worse than **EGNN** on most tasks [2509.03885]. This directly supports a nontrivial design conclusion stated in the source: topological enhancement is **not automatically beneficial** and must be architecturally integrated with care [2509.03885].

The task profile emphasizes fold classification. The reported gains are largest where **secondary structure organization** is central, especially in the hardest fold-level split on SCOP [2509.03885]. The authors interpret this as evidence that explicit SSE-level modeling and multi-rank communication capture structural patterns that residue-only GNNs represent less effectively [2509.03885]. By contrast, inverse folding remains dominated by local residue environments, where a strong residue-level GGNN can still lead [2509.03885].

## 6. Relation to prior theory, misconceptions, and limitations

TCPNet sits at the intersection of several earlier lines of work, but it should not be conflated with them. In the general topological neural network literature, a TNN is a network on a topological input space \(X\) that evaluates finitely many bounded continuous test functions and feeds the resulting finite-dimensional vector into an ordinary neural network. Under separation and strong-separation conditions on the test family, such networks are uniformly dense in the uniformly continuous functions \(C_U(X,\mathfrak S_{\mathcal M})\) associated with a canonical uniformity [2305.16639]. That result is a genuine universal approximation theorem, but it is not the theorem proved for TCPNet in protein learning [2305.16639].

A second relevant antecedent is the topological view of classification networks as continuous maps that separate labeled subsets by sending them into disjoint embedded discs or, for softmax networks, into Voronoi cells of simplex vertices [2008.13697]. That framework emphasizes that architectural choices cannot be made independently of the topology of the underlying data and that too-early dimensionality reduction can destroy separability [2008.13697]. TCPNet is compatible with that broader perspective, but its construction is specific: it uses a combinatorial complex, explicit multi-rank incidence structure, and SE(3)-equivariant scalar–vector processing tailored to proteins [2509.03885].

A common misconception is to read “topology-complete” in TCPNet as a formal completeness or universality guarantee. The Topotein source explicitly does not present the term in that sense. It states that the term is a **design philosophy**: the model uses a combinatorial complex that includes all biologically relevant ranks \(0\)–\(3\) and neighborhoods enabling communication within and across those ranks, and the TCP module is rank-agnostic in the sense that it can process scalar and vector features at any rank given appropriate frames [2509.03885]. The phrase therefore refers to architectural coverage of the hierarchy, not to a theorem that all target functions or all protein tasks are representable.

The reported limitations are correspondingly practical rather than foundational. The PCC and outer-edge neighborhoods are computationally heavier than simple residue graphs; the construction depends on reliable 3D coordinates and DSSP assignments; improvements over strong GGNN baselines are typically in the \(1\)–\(5\%\) range rather than orders of magnitude; and the method assumes a single static structure, without explicit handling of disordered regions or multiple conformations [2509.03885]. The paper also identifies future directions: integration with large sequence or structure models such as ESM3, extension to protein complexes and other biomolecular systems, improved theoretical analysis of expressivity and invariance on combinatorial complexes, and richer hierarchy designs such as adding domains as an intermediate rank [2509.03885].

Taken together, these works place TCPNet in a precise position. It is neither merely a renamed GGNN nor a generic topological universal approximator. It is a **hierarchical, SE(3)-equivariant topological neural network on a protein combinatorial complex**, designed to represent residues, contacts, SSEs, and proteins within a single message-passing system, and it is empirically strongest on tasks where **secondary-structure arrangement** and **multi-scale structural organization** are decisive [2509.03885].

Source: https://www.emergentmind.com/topics/topology-complete-perceptron-network-tcpnet