by Nathan Pinsker

How do we tell if a position can explain spottiness? Let's turn that question around and ask the reverse: how do we tell if a position can't explain spottiness? The answer to this question is that if we come across two cows, one of which is spotted and one of which is not, that have the same base in their genome at that position, then that position isn't sufficient. This is because we won't be able to tell those two cows apart.

The way to solve this problem is to check each set of three positions individually to see whether they're sufficient to explain spottiness. For each of the $O(M^3)$ possible positions, we check whether there's any matching set of bases that appears at that position in both a spotty and a non-spotty cow.

If we're considering a set of positions $(i, j, k)$, we first iterate over every non-spotty cow, and recording whether we can find an A, C, G, or T in each of the positions $i$, $j$, and $k$. We do the same for the spotty cows, and check at each step if we've already found a non-spotty cow with the same set of three bases. If we find a spotty cow that shares the same set of three bases with a non-spotty cow in this position, then these three positions don't comprise a valid candidate for explaining spottiness. If we can't find any of these overlaps, then this set of three positions can successfully be used.

The most difficult part of this problem might be keeping track of which sets of three bases we've already seen. Luckily, there's a trick to make it a lot easier! We can convert 'A' to 0, 'C' to 1, 'G' to 2, and 'T' to 3. Then we can compare two positions $(i_1, j_1, k_1)$ and $(i_2, j_2, k_2)$ by comparing the values of $16 \cdot i_1 + 4 \cdot j_1 + k_1$ and $16 \cdot i_2 + 4 \cdot j_2 + k_2$. The positions will be equal to each other if and only if these values will be equal -- for the same reason that you can compare two numbers in base 4 by comparing each of their digits.

The total runtime is $O(NM^3)$, which is fast enough to receive full credit.

Here's Brian Dean's code:

#include <iostream>
#include <fstream>
#include <cmath>
using namespace std;

int N, M;
string spotty[500], plain[500];
int S[500][50], P[500][50], A[64];

bool test_location(int j1, int j2, int j3)
  bool good = true;
  for (int i=0; i<N; i++) 
    A[S[i][j1]*16 + S[i][j2]*4 + S[i][j3]] = 1;
  for (int i=0; i<N; i++) 
    if (A[P[i][j1]*16 + P[i][j2]*4 + P[i][j3]]) good = false;
  for (int i=0; i<N; i++) 
    A[S[i][j1]*16 + S[i][j2]*4 + S[i][j3]] = 0;
  return good;

int main(void)
  ifstream fin ("");
  ofstream fout ("cownomics.out");
  fin >> N >> M;
  for (int i=0; i<N; i++) {
    fin >> spotty[i];
    for (int j=0; j<M; j++) {
       if (spotty[i][j]=='A') S[i][j] = 0;
       if (spotty[i][j]=='C') S[i][j] = 1;
       if (spotty[i][j]=='G') S[i][j] = 2;
       if (spotty[i][j]=='T') S[i][j] = 3;
  for (int i=0; i<N; i++) {
    fin >> plain[i];
    for (int j=0; j<M; j++) {
       if (plain[i][j]=='A') P[i][j] = 0;
       if (plain[i][j]=='C') P[i][j] = 1;
       if (plain[i][j]=='G') P[i][j] = 2;
       if (plain[i][j]=='T') P[i][j] = 3;
  int answer = 0;
  for (int j1=0; j1<M; j1++) 
    for (int j2=j1+1; j2<M; j2++) 
      for (int j3=j2+1; j3<M; j3++) 
	if (test_location(j1,j2,j3)) answer++;
  fout << answer << "\n";
  return 0;