HCLS AI Factory — Intelligence Report¶

Patient: GEN-2026-0087 | Pipeline Run: HCLS-VCP-2026-0087 | Mode: Full (Stages 1→2→3)

License: Apache 2.0 | Author: Adam Jones | Date: February 2026

Report Header¶

Field	Value
Patient ID	GEN-2026-0087
Run ID	HCLS-VCP-2026-0087
Pipeline Version	HLS-Pipeline v1.0.0
Pipeline Mode	Full (Genomics → RAG/Chat → Drug Discovery)
Hardware	NVIDIA DGX Spark (GB10 GPU, 128 GB unified)
Total Pipeline Duration	4 hours 12 minutes
Report Generated	February 2026
Status	COMPLETE — 100 novel drug candidates ranked

1. Genomics Summary (Stage 1)¶

Input Data¶

Parameter	Value
Sample	HG002 (NA24385, GIAB Ashkenazi male)
Sequencing	Illumina, 30× WGS, 2×250 bp paired-end
FASTQ Size	198.7 GB (R1: 99.4 GB, R2: 99.3 GB)
Reference Genome	GRCh38 (3.1 GB)

Parabricks Execution¶

Step	Tool	Duration	GPU Utilization	Peak Memory
Alignment	BWA-MEM2 (fq2bam)	34 min	82%	38 GB
Variant Calling	Google DeepVariant	22 min	91%	54 GB
Total Stage 1		56 min

VCF Output¶

Metric	Count
Total Variants Called	11,724,891
PASS Quality (QUAL>30)	3,487,216
SNPs	4,198,433
Indels	1,012,548
Multi-allelic Sites	148,762
Coding Region Variants	35,616
Transition/Transversion Ratio	2.07

Quality Assessment: PASS — Ts/Tv ratio within expected range (2.0-2.1), variant counts consistent with 30× WGS of Ashkenazi ancestry sample.

2. Annotation & Target Identification (Stage 2)¶

Annotation Funnel¶

Stage	Variants	Filter Applied
Raw VCF	11,724,891	—
Quality Filter	3,487,216	QUAL > 30
ClinVar Annotated	35,616	Clinical significance match
AlphaMissense Scored	6,831	AI pathogenicity prediction
HIGH Impact + Pathogenic	2,412	Actionable subset
Druggable Gene Targets	847	Knowledge base match

Top 5 Target Hypotheses (Claude RAG Analysis)¶

Rank	Gene	Variant	ClinVar	AlphaMissense	Therapeutic Area	Druggability
1	VCP	rs188935092 (chr9:35065263 G>A)	Pathogenic	0.87	Neurology	0.92
2	EGFR	rs121913229 (chr7:55259515 T>G)	Pathogenic	0.79	Oncology	0.95
3	BRCA1	rs80357914 (chr17:43091434 G>A)	Pathogenic	0.72	Oncology	0.78
4	PCSK9	rs11591147 (chr1:55505647 T>G)	Pathogenic	0.68	Cardiovascular	0.88
5	CFTR	rs75527207 (chr7:117559590 G>A)	Pathogenic	0.81	Respiratory	0.71

Primary Target Selection: VCP¶

Parameter	Value
Gene	VCP (Valosin-Containing Protein / p97)
UniProt	P55072
Function	AAA+ ATPase, ubiquitin-proteasome pathway
Diseases	Frontotemporal Dementia (FTD), ALS, IBMPFD
Variant	rs188935092 — missense, HIGH impact
ClinVar Classification	Pathogenic (reviewed by expert panel)
AlphaMissense Score	0.87 (pathogenic, >0.564 threshold)
Druggability Score	0.92 (D2 ATPase domain, ~450 Å³ pocket)
Known Inhibitors	CB-5083 (Phase I), NMS-873
Confidence	HIGH — multiple independent evidence sources

Evidence Chain¶

Genomic: rs188935092 detected at chr9:35065263 (G>A), heterozygous, QUAL=892
Clinical: ClinVar classifies as Pathogenic for FTD/ALS/IBMPFD (expert panel review)
AI Prediction: AlphaMissense score 0.87 (pathogenic threshold >0.564)
Functional: VEP annotates as missense_variant, HIGH impact, D2 ATPase domain
Druggability: Known drug target — CB-5083 reached Phase I clinical trial
Structural: 4 PDB structures available including inhibitor-bound 5FTK

Milvus Vector Search Statistics¶

Parameter	Value
Total Embeddings Indexed	3,487,216
Embedding Model	BGE-small-en-v1.5 (384-dim)
Index Type	IVF_FLAT (nlist=1024)
Search nprobe	16
Query Expansion	Neurology therapeutic area keywords
Top-k Retrieved	20 variant contexts
Search Latency	12 ms

3. Drug Discovery Results (Stage 3)¶

Structure Evidence¶

PDB ID	Resolution	Method	Description	Score
5FTK	2.3 Å	X-ray	VCP D2 + CB-5083 inhibitor	13.2 (selected)
7K56	2.5 Å	Cryo-EM	VCP complex	10.8
8OOI	2.9 Å	Cryo-EM	WT VCP hexamer	8.9
9DIL	3.2 Å	Cryo-EM	Mutant VCP	7.4

Selected Structure: 5FTK — inhibitor-bound (CB-5083), X-ray at 2.3 Å resolution. Binding site in D2 ATPase domain, key residues: ALA464, GLY479, ASP320, GLY215.

Molecule Generation (MolMIM)¶

Parameter	Value
Seed Compound	CB-5083 (ATP-competitive VCP inhibitor)
NIM Endpoint	MolMIM (port 8001)
Molecules Generated	100
Chemically Valid	98 (2 rejected by RDKit valence check)
Generation Time	2 min 14 sec

Drug-Likeness Profile¶

Metric	Pass	Fail	Pass Rate
Lipinski Rule of Five	87	11	88.8%
QED > 0.67 (drug-like)	72	26	73.5%
QED > 0.49 (moderate+)	91	7	92.9%
TPSA < 140 Å²	94	4	95.9%

Molecular Docking (DiffDock)¶

Parameter	Value
NIM Endpoint	DiffDock (port 8002)
Protein Target	5FTK (VCP D2 domain)
Candidates Docked	98 (all valid molecules)
Docking Time	8 min 42 sec
Mean Dock Score	-7.4 kcal/mol
Best Dock Score	-11.4 kcal/mol
Scores < -8.0 (excellent)	34 candidates
Scores < -6.0 (good+)	78 candidates

Composite Scoring¶

Formula: 30% Generation + 40% Docking + 30% QED

Docking normalization: max(0, min(1, (10 + dock_score) / 20))

Top 10 Ranked Candidates¶

Rank	Composite	Gen Score	Dock (kcal/mol)	QED	MW (Da)	LogP	Lipinski
1	0.89	0.92	-11.4	0.81	423.5	3.2	PASS
2	0.86	0.88	-10.8	0.79	441.2	3.7	PASS
3	0.84	0.85	-10.2	0.82	398.7	2.9	PASS
4	0.82	0.91	-9.8	0.74	467.1	4.1	PASS
5	0.81	0.83	-9.5	0.78	412.3	3.4	PASS
6	0.79	0.87	-9.1	0.71	455.8	3.8	PASS
7	0.78	0.80	-8.9	0.76	389.2	2.7	PASS
8	0.76	0.84	-8.7	0.69	478.4	4.3	PASS
9	0.75	0.79	-8.5	0.73	401.6	3.1	PASS
10	0.74	0.82	-8.2	0.68	448.9	3.9	PASS

CB-5083 Seed Comparison¶

Metric	CB-5083 (Seed)	Top Candidate	Improvement
Dock Score	-8.1 kcal/mol	-11.4 kcal/mol	+41% binding affinity
QED	0.62	0.81	+31% drug-likeness
MW	487.2 Da	423.5 Da	-13% (better oral absorption)
Composite	0.64	0.89	+39% overall

4. Pipeline Performance¶

Stage Timing¶

Stage	Duration	GPU Util (avg)	Peak Memory
1 — Genomics (fq2bam)	34 min	82%	38 GB
1 — Genomics (DeepVariant)	22 min	91%	54 GB
2 — Annotation	18 min	15% (CPU-bound)	12 GB
2 — Milvus Indexing	24 min	35%	22 GB
2 — RAG/Chat (interactive)	45 min	5%	8 GB
3 — Structure Retrieval	2 min	0% (network I/O)	2 GB
3 — MolMIM Generation	2 min 14 sec	78%	18 GB
3 — DiffDock Docking	8 min 42 sec	85%	24 GB
3 — Scoring + Reporting	1 min 30 sec	0% (CPU)	4 GB
Total	~4 hr 12 min

Service Health at Report Generation¶

Service	Port	Status
Landing Page	8080	HEALTHY
Genomics Portal	5000	HEALTHY
Milvus	19530	HEALTHY
RAG API	5001	HEALTHY
Streamlit Chat	8501	HEALTHY
MolMIM NIM	8001	HEALTHY
DiffDock NIM	8002	HEALTHY
Discovery UI	8505	HEALTHY
Grafana	3000	HEALTHY
Prometheus	9099	HEALTHY

5. Clinical Interpretation¶

Summary¶

Patient GEN-2026-0087 carries a heterozygous pathogenic missense variant (rs188935092) in the VCP gene. This variant is associated with Frontotemporal Dementia (FTD), ALS, and Inclusion Body Myopathy with Paget Disease and Frontotemporal Dementia (IBMPFD). The variant is classified as Pathogenic by ClinVar expert panel review and scores 0.87 on the AlphaMissense pathogenicity scale.

Drug Discovery Outcome¶

The AI-driven drug discovery pipeline identified 100 novel VCP inhibitor candidates with the top candidate showing a 39% improvement in composite score over the CB-5083 seed compound. All top 10 candidates pass Lipinski's Rule of Five and show favorable QED scores (>0.67), suggesting oral drug-likeness.

Recommended Actions¶

Genetic counseling for FTD/ALS risk assessment
Experimental validation of top 5 candidates in VCP ATPase activity assays
ADMET profiling for lead optimization
Cross-modal follow-up with Imaging Intelligence Agent for neurological assessment

6. Provenance¶

Item	Value
Pipeline	HLS-Pipeline v1.0.0 (Nextflow DSL2)
Parabricks	nvcr.io/nvidia/clara/clara-parabricks:4.6.0-1
DeepVariant	Google DeepVariant (via Parabricks, >99% accuracy)
Reference	GRCh38 (3.1 GB)
ClinVar	February 2026 release (4.1M variants)
AlphaMissense	v1.0 (71,697,560 predictions)
VEP	Ensembl VEP (GRCh38)
Milvus	v2.4 (IVF_FLAT, nlist=1024, COSINE)
Embedding Model	BGE-small-en-v1.5 (384-dim)
LLM	claude-sonnet-4-20250514 (temperature=0.3)
MolMIM	nvcr.io/nvidia/clara/bionemo-molmim:1.0
DiffDock	nvcr.io/nvidia/clara/diffdock:1.0
Hardware	NVIDIA DGX Spark (GB10, 128 GB unified)
Scoring	30% gen + 40% dock + 30% QED

This is a demonstration intelligence report generated by the HCLS AI Factory. All patient data is synthetic. The report illustrates the platform's end-to-end capability from genomic variant calling through drug candidate ranking.

HCLS AI Factory — Apache 2.0 | Author: Adam Jones | February 2026