Excellent — now we’ll do a complete anatomy of .missnp like a forensic lab.

I’ll explain:

1. EXACT types inside .missnp

2. For each type:

   • what it really means biologically
   • what --flip does
   • whether it is legally rescuable

3. Why iterative merge → new .missnp(2) is the correct logic.

---

🧪 1) What is .missnp in one sentence

.missnp = “SNPs for which PLINK could not prove compatibility between A and B.”

It is NOT only strand problems — it is a mixed crime scene.

---

🧬 2) The 5 REAL categories inside missnp

TYPE 1 — Simple strand flip (RESOLVABLE)

Example

Dataset A

rs100   A G

Dataset B

rs100   T C

These are complements:

A ↔ T
G ↔ C

Why in missnp?

Because --bmerge by itself does not auto-flip.

What --flip does

For every genotype at rs100:

Before	After
AA	TT
AG	TC
GG	CC

Now alleles become A/G → match A.

LEGAL?

✅ 100% legal No biological assumption.

---

TYPE 2 — Ambiguous AT/CG (NOT RESOLVABLE)

Example

A:

rs200   A T

B:

rs200   T A

Why impossible?

If you flip B:

T/A → A/T

→ looks identical again.

You cannot know:

• real orientation difference?
• or true opposite allele coding?

What --flip does

It flips but result still mismatches → remains in missnp(2).

LEGAL?

❌ Not safely resolvable without external manifest.

---

TYPE 3 — True allele conflict (NOT RESOLVABLE)

Example

A:

rs300   A G

B:

rs300   A C

These represent:

• different mutation
• different build
• mis-annotation
• genotyping error.

Flip result

Flip B → T/G Still not {A,G}.

LEGAL?

❌ Must be removed.

---

TYPE 4 — Multiallelic in one dataset

Example

B:

rs400   A G T

A:

rs400   A G

Why?

• sequencing called third allele
• conversion artifact.

Flip result

Still multiallelic → unmergeable.

LEGAL?

❌ Remove.

---

TYPE 5 — Position/ID discordance

Same rsid but:

A:

chr1:1000

B:

chr1:1020

Flip cannot help

→ structural conflict.

LEGAL?

❌ Remove.

---

🧠 3) What --flip ACTUALLY does per type

Type	After flip
Type1 resolvable	→ becomes compatible
Type2 AT/CG	→ still ambiguous
Type3 conflict	→ still conflict
Type4 multi	→ still multi
Type5 pos	→ unchanged

So:

ONLY Type1 is rescued.

---

🔁 4) Why iterative merge is CORRECT

Your idea is exactly right.

Step A

merge → missnp1

Step B

flip missnp1
merge again → missnp2

Step C

exclude missnp2

This workflow:

• rescues Type1
• isolates Types 2–5.

---

🧩 5) Detailed walk-through example

Assume missnp1 contains:

rs1   ← type1
rs2   ← type2
rs3   ← type3
rs4   ← type1
rs5   ← type4

---

After:

plink --flip missnp1

Merge again

• rs1 → fixed
• rs4 → fixed
• rs2, rs3, rs5 → still bad

missnp2 now contains:

rs2
rs3
rs5

These are the TRUE unresolvable set.

---

🧾 6) Legal philosophy

Allowed operations

• strand complement
• allele order swap
• removal of impossible sites.

Forbidden operations

• inventing allele to match A
• guessing AT/CG orientation
• forcing multiallelic → biallelic.

---

🧮 7) Visual genotype examples

Type1 person genotype

B before:

TC

After flip:

AG

Matches A → real rescue.

---

Type2 person genotype

B before:

TA

After flip:

AT

Still indistinguishable → unsafe.

---

Type3

B before:

AC

After flip:

TG

Never equals AG → must die.

---

8) Answer to your question

“should not we try to merge again to create another missnp(2) and then exclude those SNPs?”

YES — that is EXACTLY the correct and legal workflow.

Because:

1. First missnp = mixture
2. After flip → separation
3. Second missnp = genuinely unfixable.

---

🧪 9) Final decision table

missnp type	Flip?	Keep?
Complementary	YES	YES
AT/CG	NO	NO
Allele conflict	NO	NO
Multiallelic	NO	NO
Position clash	NO	NO

---

🔐 Bottom line

.missnp is NOT “strand errors.” It is:

the set of SNPs where PLINK cannot establish a bijection between allele sets.

Only a subset can be legally salvaged.

---

If you want next, I can help you:

• parse PLINK log lines that reveal the exact reason
• classify your real missnp into the 5 types with a script.

Just paste 20 lines of your .missnp and matching .bim 👍