File size: 8,154 Bytes
9f5e507
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
"""
SBModel — Anisotropic Schrödinger Bridge model.

Shared backbone with scDFM, dual output heads (velocity + score),
plus AnisotropicSigmaNet for per-gene diffusion coefficients.
"""

import torch
import torch.nn as nn
from torch import Tensor
from typing import Optional, Tuple

from .layers import AnisotropicSigmaNet, ScoreDecoder
from .._scdfm_imports import (
    GeneadaLN,
    ContinuousValueEncoder,
    GeneEncoder,
    BatchLabelEncoder,
    TimestepEmbedder,
    ExprDecoder,
    DifferentialTransformerBlock,
    PerceiverBlock,
    DiffPerceiverBlock,
)


class SBModel(nn.Module):
    """
    Anisotropic Schrödinger Bridge model.

    forward(gene_id, cell_1, x_t, t, perturbation_id)
      → (pred_velocity, pred_score, sigma_g)

    - pred_velocity: (B, G) PF-ODE velocity (target = x_T - x₀)
    - pred_score:    (B, G) score function (target = conditional score)
    - sigma_g:       (B, G) per-gene diffusion coefficient in [σ_min, σ_max]
    """

    def __init__(
        self,
        ntoken: int = 6000,
        d_model: int = 128,
        nhead: int = 8,
        d_hid: int = 512,
        nlayers: int = 4,
        dropout: float = 0.1,
        fusion_method: str = "differential_perceiver",
        perturbation_function: str = "crisper",
        use_perturbation_interaction: bool = True,
        mask_path: str = None,
        # Sigma net params
        sigma_min: float = 0.01,
        sigma_max: float = 2.0,
        sigma_init: float = 0.5,
        sigma_hidden_dim: int = 256,
        sigma_num_layers: int = 2,
        # Score decoder params
        score_head_depth: int = 2,
        use_score: bool = True,
    ):
        super().__init__()
        self.d_model = d_model
        self.fusion_method = fusion_method
        self.perturbation_function = perturbation_function
        self.use_score = use_score

        # === Timestep embedder (single, not cascaded) ===
        self.t_embedder = TimestepEmbedder(d_model)

        # === Perturbation embedder ===
        self.perturbation_embedder = BatchLabelEncoder(ntoken, d_model)

        # === Expression stream (reused from scDFM) ===
        self.value_encoder_1 = ContinuousValueEncoder(d_model, dropout)
        self.value_encoder_2 = ContinuousValueEncoder(d_model, dropout)
        self.encoder = GeneEncoder(
            ntoken, d_model,
            use_perturbation_interaction=use_perturbation_interaction,
            mask_path=mask_path,
        )

        self.fusion_layer = nn.Sequential(
            nn.Linear(2 * d_model, d_model),
            nn.GELU(),
            nn.Linear(d_model, d_model),
            nn.LayerNorm(d_model),
        )

        # === Shared backbone blocks ===
        if fusion_method == "differential_transformer":
            self.blocks = nn.ModuleList([
                DifferentialTransformerBlock(d_model, nhead, i, mlp_ratio=4.0)
                for i in range(nlayers)
            ])
        elif fusion_method == "differential_perceiver":
            self.blocks = nn.ModuleList([
                DiffPerceiverBlock(d_model, nhead, i, mlp_ratio=4.0)
                for i in range(nlayers)
            ])
        elif fusion_method == "perceiver":
            self.blocks = nn.ModuleList([
                PerceiverBlock(d_model, d_model, heads=nhead, mlp_ratio=4.0, dropout=0.1)
                for _ in range(nlayers)
            ])
        else:
            raise ValueError(f"Invalid fusion method: {fusion_method}")

        # === Per-layer gene AdaLN + adapter ===
        self.gene_adaLN = nn.ModuleList([
            GeneadaLN(d_model, dropout) for _ in range(nlayers)
        ])
        self.adapter_layer = nn.ModuleList([
            nn.Sequential(
                nn.Linear(2 * d_model, d_model),
                nn.LeakyReLU(),
                nn.Dropout(dropout),
                nn.Linear(d_model, d_model),
                nn.LeakyReLU(),
            )
            for _ in range(nlayers)
        ])

        # === Velocity decoder head (reused ExprDecoder from scDFM) ===
        self.final_layer = ExprDecoder(d_model, explicit_zero_prob=False, use_batch_labels=True)

        # === Score decoder head (NEW) ===
        if use_score:
            self.score_decoder = ScoreDecoder(d_model, depth=score_head_depth)

        # === Anisotropic sigma network (NEW, independent of backbone) ===
        self.sigma_net = AnisotropicSigmaNet(
            d_model=d_model,
            hidden_dim=sigma_hidden_dim,
            num_layers=sigma_num_layers,
            sigma_min=sigma_min,
            sigma_max=sigma_max,
            sigma_init=sigma_init,
        )

        self.initialize_weights()

    def initialize_weights(self):
        def _basic_init(module):
            if isinstance(module, nn.Linear):
                torch.nn.init.xavier_uniform_(module.weight)
                if module.bias is not None:
                    nn.init.constant_(module.bias, 0)
        self.apply(_basic_init)
        # Re-initialize sigma bias after global init
        self.sigma_net._init_bias(self.sigma_net.sigma_min +
            (self.sigma_net.sigma_max - self.sigma_net.sigma_min) * 0.5)

    def get_perturbation_emb(
        self,
        perturbation_id: Optional[Tensor] = None,
        perturbation_emb: Optional[Tensor] = None,
        cell_1: Optional[Tensor] = None,
    ) -> Tensor:
        """Get perturbation embedding, replicating scDFM logic."""
        assert perturbation_emb is None or perturbation_id is None
        if perturbation_id is not None:
            if self.perturbation_function == "crisper":
                perturbation_emb = self.encoder(perturbation_id)
            else:
                perturbation_emb = self.perturbation_embedder(perturbation_id)
            perturbation_emb = perturbation_emb.mean(1)
        elif perturbation_emb is not None:
            perturbation_emb = perturbation_emb.to(cell_1.device, dtype=cell_1.dtype)
            if perturbation_emb.dim() == 1:
                perturbation_emb = perturbation_emb.unsqueeze(0)
            if perturbation_emb.size(0) == 1:
                perturbation_emb = perturbation_emb.expand(cell_1.shape[0], -1).contiguous()
            perturbation_emb = self.perturbation_embedder.enc_norm(perturbation_emb)
        return perturbation_emb

    def forward(
        self,
        gene_id: Tensor,             # (B, G) gene token IDs
        cell_1: Tensor,               # (B, G) source expression
        x_t: Tensor,                  # (B, G) noised target expression
        t: Tensor,                    # (B,) timestep
        perturbation_id: Optional[Tensor] = None,
    ) -> Tuple[Tensor, Optional[Tensor], Tensor]:
        if t.dim() == 0:
            t = t.repeat(cell_1.size(0))

        # 1. Expression embedding (aligned with scDFM)
        gene_emb = self.encoder(gene_id)                         # (B, G, d)
        val_emb_1 = self.value_encoder_1(x_t)
        val_emb_2 = self.value_encoder_2(cell_1) + gene_emb
        x = self.fusion_layer(torch.cat([val_emb_1, val_emb_2], dim=-1)) + gene_emb

        # 2. Conditioning vector (single t, no cascaded)
        t_emb = self.t_embedder(t)
        pert_emb = self.get_perturbation_emb(perturbation_id, cell_1=cell_1)
        c = t_emb + pert_emb

        # 3. Shared backbone
        for i, block in enumerate(self.blocks):
            x = self.gene_adaLN[i](gene_emb, x)
            pert_exp = pert_emb[:, None, :].expand(-1, x.size(1), -1)
            x = torch.cat([x, pert_exp], dim=-1)
            x = self.adapter_layer[i](x)
            x = block(x, val_emb_2, c)

        # 4a. Velocity head
        x_with_pert = torch.cat([x, pert_emb[:, None, :].expand(-1, x.size(1), -1)], dim=-1)
        pred_velocity = self.final_layer(x_with_pert)["pred"]   # (B, G)

        # 4b. Score head
        pred_score = None
        if self.use_score:
            pred_score = self.score_decoder(x, pert_emb)         # (B, G)

        # 4c. Sigma (independent of backbone, only depends on pert_emb, t, gene_emb)
        sigma_g = self.sigma_net(pert_emb, t, gene_emb)          # (B, G)

        return pred_velocity, pred_score, sigma_g