Fragment Analysis

Overview

The Fragment Analysis tab implements random masking AAS attribution and downstream positive-fragment workflows. Main capabilities:

• Per-candidate attribution: Monte Carlo masking (random fraction or connected subgraph) or threshold baseline (CSS-style τ − |Δδ|) on Top-N screening hits.
• Structure coloring: 2D structure with per-carbon mean attribution scores; optional atom index and predicted δ labels; HD export (SVG/PNG).
• Positive fragment extraction: Threshold (manual or robust Z), bridging rules, fragment–spectrum matching with DEPT/greedy options.
• Intra-molecular fusion: Merge top fragments within one candidate.
• Cross-candidate fusion: Top fragment combinations across the analysis pool.
• Global fragment ranking: Table and plot of all extracted fragments sorted by Σscore.
• AI narrative: AI: Fragment attribution narrative (requires LLM configuration).

Prerequisite: completed Molecular Analysis with at least one hit and available experimental peaks.

Interface Layout

Left panel (collapsible sections, max width ~650 px)

• Random masking AAS — Monte Carlo parameters
• Positive fragment extraction (threshold, bridging, extract button, intra-molecular fusion)
• Fragment fusion (Top combinations)
• Status label, Run masking attribution, AI: Fragment attribution narrative
• Progress bar during batch masking

Right panel

• Result view combo: Compound attribution / Fusion combinations
• Attribution structure image + per-carbon score table
• Fragment table (per candidate); Global fragment table + structure scroll
• Fusion table and inline preview card; optional NMR mapping canvases

Workflow

① Prerequisites on Molecular Analysis tab

• Run Start Analysis and obtain hits in Result.csv.
• Ensure experimental peaks exist:
• Manual peak input, or
• GUI_result_files/NMR-1D.csv after preprocessing.

② Set masking parameters and run attribution

Expand Random masking AAS — Monte Carlo parameters.

Parameter	Default	Description
Top N candidates to analyze	10	First N rows from screening results
Monte Carlo mask mode	Random connected	See table below
Connected mask size k	3	Carbons per connected mask (connected mode only)
Random subset fraction f	0.5	Fraction mode: ⌈f·N⌉ carbons per iteration
Threshold τ (ppm)	3.00	Threshold-baseline mode only
Monte Carlo iterations R	300	MC iterations (not used in threshold mode)
Random seed	empty	Integer for reproducibility; empty = random
Consider DEPT type matching	✓	Match q/t/d/s between prediction and experiment
Greedy unique peak matching	✓	Greedy nearest peak, remove matched peaks
Shuffle virtual δ order...	✓	With greedy: shuffle prediction order before each AAS
Fixed ±T (cross-molecule)	✓	Shared color scale ±T
T	5.0	Half-range for red–green color map

Mask modes

Mode	Behavior
Random fraction	Each iteration masks ⌈f·N⌉ arbitrary carbons
Random connected subgraph	Each iteration masks k bonded carbons (default)
Threshold baseline (CSS)	Deterministic τ − |Δδ|; no MC, R, f, k, or seed

• Click Run masking attribution. Top N candidates are processed sequentially in the background; combo box lists completed candidates.

• Apply color scale only (no re-run MC): Recolor structure after changing T or fixed-scale option.

③ Extract positive fragments

Expand Positive fragment extraction.

Parameter	Default	Description
Threshold mode	Manual (>T)	Or robust Z: median + k·MAD
Score threshold (>T)	0.0	Manual cutoff on mean scores
Robust Z k	1.0	Used in robust Z mode
Min carbons in fragment	2	Minimum fragment size
Bridge via low-score carbons	optional	Merge clusters through low-score nodes
Allow hetero bridge (C-X-C)	✓	Connect across O/N/S, etc.
Max bridge low-score carbons	1	Bridge path length limit
Fragment ↔ spectrum: DEPT	✓	Type-matched fragment–peak matching
Fragment ↔ spectrum: greedy unique	✓	Independent from masking MC settings
Greedy shuffle repeats	100	≥2: keep best Σ(Δδ)² over shuffles
Max fragments listed	50	Cap on fragment table rows

• Click Extract positive fragments + map.
• Click a fragment row to open the NMR matching map dialog.

④ Optional: Intra-molecular fusion

Parameter	Default
Pool Top fragments	30
Max merged fragments	5
Greedy shuffle repeats	10
Coverage weight λ	0.3

• Run intra-molecular fusion (current candidate) — uses fragments from the selected candidate only.

⑤ Optional: Fragment fusion (Top combinations)

Parameter	Default
Fusion pool Top fragments	30
Fusion max merged fragments	5
Fusion greedy shuffle repeats	10
Fusion coverage weight λ	0.3

• Run fusion analysis (Top combinations).
• Switch Result view to Fusion combinations view; click fusion rows for detail / spectrum maps.

⑥ Review structures and export

• Toggle Show RDKit atom index, Show predicted δ (ppm).
• Rotate / Flip H / Flip V — preview updates live.
• Save HD (SVG/PNG) — exports current attribution view; may write *_analysis_params.json sidecar.

⑦ Output and in-memory artifacts

No fixed extra folder beyond GUI exports you trigger. Key artifacts:

• Colored structure images (user-selected path)
• Fragment / fusion tables on screen
• Context for AI Assistant fragment tasks

FAQ

1. Run masking attribution disabled

Complete Start Analysis on Molecular Analysis with ≥1 hit.

1. No experimental peaks

Provide manual peaks or NMR-1D.csv.

1. Extract / fusion buttons disabled

Finish masking for at least one candidate first.

1. Threshold baseline vs MC options grayed out

UI enables f, k, τ, R, and seed depending on selected mask mode.

1. Colors look wrong across molecules

Turn on Fixed ±T and adjust T, then Apply color scale only.

1. Empty global fragment table

Run Extract positive fragments for multiple candidates in the Top-N batch.

Relation to Other Modules

• Molecular Analysis: Supplies Result.csv, SMILES, predicted shifts/types, and experimental peaks — required.
• AI Assistant: Compare masking, Global consensus, Candidate attr., fragment table context menu; needs masking (and preferably extraction) data here.
• Database Management: No direct link.
• Other Tool: Independent structure drawing.