-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathlearning_pack.html
More file actions
285 lines (252 loc) · 14.1 KB
/
Copy pathlearning_pack.html
File metadata and controls
285 lines (252 loc) · 14.1 KB
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
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>ZRT BioNeMo 5-Lesson Learning Pack</title>
<!-- Google Fonts -->
<link href="https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700&display=swap" rel="stylesheet">
<style>
:root {
--bg-color: #0b0f19;
--card-bg: rgba(17, 24, 39, 0.7);
--border-color: rgba(255, 255, 255, 0.08);
--primary: #00f2fe;
--primary-rgb: 0, 242, 254;
--secondary: #4facfe;
--text-main: #f3f4f6;
--text-muted: #9ca3af;
--success: #10b981;
--warning: #f59e0b;
}
* { box-sizing: border-box; margin: 0; padding: 0; }
body {
font-family: 'Inter', sans-serif;
background-color: var(--bg-color);
background-image:
radial-gradient(at 0% 0%, rgba(79, 172, 254, 0.1) 0px, transparent 50%),
radial-gradient(at 100% 0%, rgba(0, 242, 254, 0.08) 0px, transparent 50%);
background-attachment: fixed;
color: var(--text-main);
line-height: 1.6;
}
header {
border-bottom: 1px solid var(--border-color);
backdrop-filter: blur(12px);
-webkit-backdrop-filter: blur(12px);
padding: 1.5rem 2rem;
display: flex;
justify-content: space-between;
align-items: center;
background: rgba(11, 15, 25, 0.6);
position: sticky;
top: 0;
z-index: 100;
}
.logo-container { display: flex; align-items: center; gap: 0.75rem; text-decoration: none;}
.logo-icon {
width: 2rem; height: 2rem;
background: linear-gradient(135deg, var(--primary), var(--secondary));
border-radius: 6px;
display: flex; align-items: center; justify-content: center;
font-weight: 700; color: #0b0f19;
box-shadow: 0 0 20px rgba(0, 242, 254, 0.3);
}
.page-nav {
display: flex;
gap: 0.75rem;
align-items: center;
flex-wrap: wrap;
}
.badge {
display: inline-flex;
align-items: center;
gap: 0.35rem;
padding: 0.45rem 0.8rem;
border-radius: 9999px;
border: 1px solid var(--border-color);
font-size: 0.75rem;
font-weight: 500;
letter-spacing: 0.05em;
text-transform: uppercase;
color: var(--text-main);
background: rgba(255, 255, 255, 0.04);
text-decoration: none;
}
.badge:hover,
.badge:focus-visible {
outline: 2px solid #fff08a;
outline-offset: 3px;
border-color: #fff08a;
}
.badge-current {
border-color: rgba(16, 185, 129, 0.5);
color: var(--success);
background: rgba(16, 185, 129, 0.1);
}
h1.logo-title {
font-size: 1.25rem; font-weight: 600;
background: linear-gradient(to right, #ffffff, #d1d5db);
-webkit-background-clip: text; -webkit-text-fill-color: transparent;
margin: 0;
}
main {
max-width: 900px;
margin: 0 auto;
padding: 3rem 2rem;
}
.hero { text-align: center; margin-bottom: 4rem; }
.hero h1 { font-size: 2.5rem; margin-bottom: 1rem; color: #fff; }
.hero p { font-size: 1.125rem; color: var(--text-muted); }
.lesson {
background: var(--card-bg);
border: 1px solid var(--border-color);
border-radius: 16px;
padding: 2rem;
margin-bottom: 2rem;
backdrop-filter: blur(16px);
box-shadow: 0 4px 30px rgba(0, 0, 0, 0.2);
transition: border-color 0.3s ease;
}
.lesson:hover { border-color: rgba(var(--primary-rgb), 0.2); }
.lesson h2 { font-size: 1.75rem; margin-bottom: 1rem; color: var(--primary); }
.lesson h3 { font-size: 1.25rem; margin-top: 1.5rem; margin-bottom: 0.75rem; color: #fff; }
p { margin-bottom: 1rem; }
ul, ol { margin-bottom: 1rem; padding-left: 1.5rem; }
li { margin-bottom: 0.5rem; }
.alert {
background: rgba(var(--primary-rgb), 0.1);
border-left: 4px solid var(--primary);
padding: 1rem;
border-radius: 4px;
margin: 1.5rem 0;
}
.alert strong { color: var(--primary); display: block; margin-bottom: 0.25rem; }
code {
background: rgba(0, 0, 0, 0.5);
padding: 0.2rem 0.4rem;
border-radius: 4px;
font-family: monospace;
font-size: 0.9em;
color: #d1d5db;
}
pre {
background: #080b11;
border: 1px solid var(--border-color);
padding: 1rem;
border-radius: 8px;
overflow-x: auto;
margin: 1.5rem 0;
}
pre code { background: none; padding: 0; color: var(--text-main); }
.btn {
background: linear-gradient(135deg, var(--primary), var(--secondary));
border: none; color: #0b0f19; padding: 0.6rem 1.2rem;
font-size: 0.875rem; font-weight: 600; border-radius: 8px;
cursor: pointer; text-decoration: none; display: inline-block;
}
</style>
</head>
<body>
<header>
<a href="/" class="logo-container">
<div class="logo-icon">Z</div>
<h1 class="logo-title">ZRT Protein Designer</h1>
</a>
<nav class="page-nav" aria-label="Primary">
<a href="/" class="badge">Home / Lab</a>
<a href="/viewer/" class="badge">Viewer</a>
<span class="badge badge-current" aria-current="page">Learning Pack</span>
</nav>
</header>
<main>
<div class="hero">
<h1>NVIDIA BioNeMo & Protein Folding</h1>
<p>From Zero to De Novo Design. A 5-Lesson Curriculum.</p>
</div>
<section class="lesson" id="lesson-1">
<h2>Lesson 1: Protein Physics & The Folding Problem</h2>
<p>Proteins are the molecular workhorses of life, orchestrating almost every biological process. To design them, we must first understand how their physical chemistry drives their three-dimensional shapes.</p>
<h3>The Central Dogma & Amino Acids</h3>
<p>Proteins start as linear chains of <strong>amino acids</strong> translated from mRNA. Each amino acid shares a common backbone but has a unique sidechain (often designated as the <strong>R-group</strong>).</p>
<ul>
<li><strong>Hydrophobic (Apolar)</strong>: Alanine (A), Valine (V), Leucine (L), Isoleucine (I), etc.</li>
<li><strong>Polar (Uncharged)</strong>: Serine (S), Threonine (T), Glutamine (Q), etc.</li>
<li><strong>Charged</strong>: Lysine (K+, basic), Aspartic Acid (D-, acidic), etc.</li>
</ul>
<h3>Hierarchical Protein Structure</h3>
<ol>
<li><strong>Primary (1°) Structure</strong>: The linear sequence of amino acids linked by covalent peptide bonds.</li>
<li><strong>Secondary (2°) Structure</strong>: Local folding patterns stabilized by hydrogen bonds between backbone carbonyl (C=O) and amide (N-H) groups (Alpha-helices and Beta-sheets).</li>
<li><strong>Tertiary (3°) Structure</strong>: The overall 3D folding of a single polypeptide chain, driven by hydrophobic collapse.</li>
<li><strong>Quaternary (4°) Structure</strong>: The assembly of multiple folded polypeptide subunits into a functional complex.</li>
</ol>
<h3>Anfinsen's Dogma & Levinthal's Paradox</h3>
<ul>
<li><strong>Anfinsen's Dogma</strong> states that, for small globular proteins, the 3D structure is determined solely by the primary sequence and represents the thermodynamic global minimum free energy state.</li>
<li><strong>Levinthal's Paradox</strong> notes that folding by random sampling would take longer than the age of the universe. Yet, proteins fold spontaneously in milliseconds, implying folding is directed along specific local energy pathways.</li>
</ul>
</section>
<section class="lesson" id="lesson-2">
<h2>Lesson 2: Introduction to NVIDIA BioNeMo</h2>
<p>NVIDIA BioNeMo is a domain-specific cloud and software platform designed to accelerate digital biology and drug discovery. It hosts state-of-the-art models as optimized Microservices (NIMs) that can be run either via hosted API endpoints or deployed locally as Docker containers on NVIDIA GPUs.</p>
<h3>The BioNeMo NIM Architecture</h3>
<p>NIM (NVIDIA Inference Microservice) packages models with optimized runtimes (like TensorRT-LLM and Triton Inference Server) so they can execute lightning-fast forward passes.</p>
<h3>Key API Integration Patterns</h3>
<p>To interact with BioNeMo's hosted endpoints, you authenticate using an API key (<code>Authorization: Bearer $NGC_API_KEY</code>) and communicate via JSON payloads.</p>
<div class="alert">
<strong>Important</strong>
The hosted endpoint URL structure for molecular modeling is: <code>https://health.api.nvidia.com/v1/biology/{developer}/{model}/{action}</code>
</div>
</section>
<section class="lesson" id="lesson-3">
<h2>Lesson 3: Generative Backbone Design with RFDiffusion</h2>
<p>Traditionally, protein design relied on modifying existing natural proteins. <strong>De novo design</strong> allows us to generate entirely new proteins from scratch to target specific shapes or receptors.</p>
<h3>How RFDiffusion Works</h3>
<p>RFDiffusion is a generative model based on <strong>Denoising Diffusion Probabilistic Models (DDPM)</strong>. Starting from complete structural noise, it iteratively denoises the coordinate positions over a series of steps (typically 50) until it resolves into a clean, biologically viable protein backbone.</p>
<h3>The Contig DSL (Domain Specific Language)</h3>
<p>RFDiffusion uses a specialized syntax to define what sections of a protein to generate and what parts of an existing target to keep fixed:</p>
<ul>
<li><code>"100"</code>: Generate a de novo backbone of exactly 100 residues.</li>
<li><code>"A1-50/0 80"</code>: Keep residues 1 to 50 of chain A and design an 80-residue binder around it separated by a chain break.</li>
</ul>
<div class="alert">
<strong>Tip</strong>
When running de novo designs hosted on NVIDIA's API, a dummy input PDB (like a single Alanine) must be provided in the <code>input_pdb</code> payload parameter to pass endpoint validation.
</div>
</section>
<section class="lesson" id="lesson-4">
<h2>Lesson 4: Sequence Design via Inverse Folding (ProteinMPNN)</h2>
<p>Once RFDiffusion outputs a 3D backbone coordinate file, we have a shape but no sequence of amino acids to make it. <strong>ProteinMPNN</strong> solves the "inverse folding" problem.</p>
<h3>Message Passing Neural Networks (MPNN)</h3>
<p>While folding models go from Sequence → Structure, ProteinMPNN goes from <strong>Structure → Sequence</strong>. It models the protein backbone as a graph and performs message passing over this graph to predict the probability distribution of the 20 amino acids at each position.</p>
<h3>Fine-Tuning Design Parameters</h3>
<ul>
<li><strong>Sampling Temperature</strong>: Controls sequence diversity. Low temperatures yield highly deterministic results. Higher temperatures introduce sequence variation.</li>
<li><strong>Omit Amino Acids</strong>: You can explicitly block specific amino acids (e.g., avoiding Cysteines <code>"omit_AAs": ["C"]</code> to prevent unwanted disulfide cross-linking).</li>
</ul>
</section>
<section class="lesson" id="lesson-5">
<h2>Lesson 5: High-Fidelity Structure Prediction (OpenFold3 and Boltz2)</h2>
<p>To verify if our designed sequence will fold into the target backbone shape we generated, we run forward structure prediction models. This forms an in silico validation loop.</p>
<h3>OpenFold3 and Boltz2</h3>
<ul>
<li><strong>OpenFold3</strong> is a fast, PyTorch-based implementation of AlphaFold3. It predicts 3D structures and confidence scores.</li>
<li><strong>Boltz2</strong> is a highly accurate open-source biomolecular structure prediction model optimized for multi-entity complexes.</li>
</ul>
<h3>Key Validation Metrics</h3>
<p>To evaluate if a designed protein is successful, we look at the following metrics:</p>
<ul>
<li><strong>pLDDT (Predicted Local Distance Difference Test)</strong>: A per-residue confidence score from 0 to 100. ≥90 is high confidence, 70-90 is good confidence, and <50 is low confidence.</li>
<li><strong>TM-Score / RMSD (Root-Mean-Square Deviation)</strong>: Measures structural similarity between the predicted structure and the original designed backbone. An RMSD <2.0 Å indicates excellent structural agreement.</li>
</ul>
<h3>The Design Loop Summary</h3>
<ol>
<li><strong>RFDiffusion</strong> creates the coordinate envelope.</li>
<li><strong>ProteinMPNN</strong> writes the code (sequence).</li>
<li><strong>OpenFold3</strong> validates that the code compiles back into the correct shape.</li>
</ol>
</section>
</main>
</body>
</html>