static bool find_files(std::string pathname, const std::string &filename, string_array &files, bool recursive)
{
if (!pathname.empty())
{
char c = pathname[pathname.size() - 1];
if ((c != ':') && (c != '\\') && (c != '/'))
pathname += "/";
}
DIR *dp = opendir(pathname.c_str());
if (!dp)
return false;
string_array paths;
for ( ; ; )
{
struct dirent *ep = readdir(dp);
if (!ep)
break;
const bool is_directory = (ep->d_type & DT_DIR) != 0;
const bool is_file = (ep->d_type & DT_REG) != 0;
if (ep->d_name[0] == '.')
continue;
std::string filename(ep->d_name);
if (is_directory)
{
if (recursive)
paths.push_back(filename);
}
else if (is_file)
files.push_back(pathname + filename);
}
closedir(dp);
dp = NULL;
if (recursive)
{
for (uint i = 0; i < paths.size(); i++)
{
const std::string &path = paths[i];
if (!find_files(pathname + path, filename, files, true))
return false;
}
}
return true;
}
bool MyFb2File::GetGenres(string_array &genres)
{
bool retVal = false;
int pos = 0;
if (m_titleInfo.length() == 0)
{
m_titleInfo = GetXmlTag(titleInfoTag, pos);
}
if (m_titleInfo.length() == 0)
{
return retVal;
}
genres.clear();
pos = 0;
QString curGenre = tr("");
while ((curGenre = GetXmlTag(genreTag, m_titleInfo, pos)).length() > 0)
{
int comma = curGenre.indexOf(',');
if (comma != -1)
{
int pos = 0;
QString g = tr("");
while ((comma = curGenre.indexOf(',', pos)) != -1)
{
int beg = pos, end = comma;
while (curGenre.mid(beg, 1) == tr(" "))
{
beg++;
}
while (curGenre.mid(end - 1, 1) == tr(" "))
{
end--;
}
if ((end - beg) > 0)
{
g = curGenre.mid(beg, end - beg);
genres.push_back(Utf8toAnsi(g));
}
pos = comma + 1;
}
int beg = pos, end = curGenre.length();
while (curGenre.mid(beg, 1) == tr(" "))
{
beg++;
}
while (curGenre.mid(end - 1, 1) == tr(" "))
{
end--;
}
if ((end - beg) > 0)
{
g = curGenre.mid(beg, end - beg);
genres.push_back(Utf8toAnsi(g));
}
}
else
{
genres.push_back(Utf8toAnsi(curGenre));
}
retVal = true;
}
return retVal;
}
int main(int argc, char* argv[])
{
// Check the required number of command line arguments.
if (argc != 5)
{
cout << "usr host user pwd jobs_path" << endl;
return 0;
}
// Fetch command line arguments.
const auto host = argv[1];
const auto user = argv[2];
const auto pwd = argv[3];
const path jobs_path = argv[4];
// Connect to host and authenticate user.
DBClientConnection conn;
{
cout << local_time() << "Connecting to " << host << " and authenticating " << user << endl;
string errmsg;
if ((!conn.connect(host, errmsg)) || (!conn.auth("istar", user, pwd, errmsg)))
{
cerr << local_time() << errmsg << endl;
return 1;
}
}
// Initialize constants.
cout << local_time() << "Initializing" << endl;
const auto collection = "istar.usr";
const auto epoch = date(1970, 1, 1);
const size_t num_usrs = 2;
constexpr array<size_t, num_usrs> qn{{ 12, 60 }};
constexpr array<double, num_usrs> qv{{ 1.0 / qn[0], 1.0 / qn[1] }};
const size_t num_references = 4;
const size_t num_subsets = 5;
const array<string, num_subsets> SubsetSMARTS
{{
"[!#1]", // heavy
"[#6+0!$(*~[#7,#8,F]),SH0+0v2,s+0,S^3,Cl+0,Br+0,I+0]", // hydrophobic
"[a]", // aromatic
"[$([O,S;H1;v2]-[!$(*=[O,N,P,S])]),$([O,S;H0;v2]),$([O,S;-]),$([N&v3;H1,H2]-[!$(*=[O,N,P,S])]),$([N;v3;H0]),$([n,o,s;+0]),F]", // acceptor
"[N!H0v3,N!H0+v4,OH+0,SH+0,nH+0]", // donor
}};
// Initialize variables.
array<array<double, qn.back()>, 1> qw;
array<array<double, qn.back()>, 1> lw;
auto q = qw[0];
auto l = lw[0];
// Read ZINC ID file.
const string_array<size_t> zincids("16_zincid.txt");
const auto num_ligands = zincids.size();
// Read SMILES file.
const string_array<size_t> smileses("16_smiles.txt");
assert(smileses.size() == num_ligands);
// Read supplier file.
const string_array<size_t> suppliers("16_supplier.txt");
assert(suppliers.size() == num_ligands);
// Read property files of floating point types and integer types.
const auto zfproperties = read<array<float, 4>>("16_zfprop.f32");
assert(zfproperties.size() == num_ligands);
const auto ziproperties = read<array<int16_t, 5>>("16_ziprop.i16");
assert(ziproperties.size() == num_ligands);
// Open files for subsequent reading.
std::ifstream usrcat_bin("16_usrcat.f64");
stream_array<size_t> ligands("16_ligand.pdbqt");
assert(ligands.size() == num_ligands);
array<vector<double>, 2> scores
{{
vector<double>(num_ligands, 0),
vector<double>(num_ligands, 0)
}};
const auto& u0scores = scores[0];
const auto& u1scores = scores[1];
vector<size_t> scase(num_ligands);
// Enter event loop.
cout << local_time() << "Entering event loop" << endl;
bool sleeping = false;
while (true)
{
// Fetch an incompleted job in a first-come-first-served manner.
if (!sleeping) cout << local_time() << "Fetching an incompleted job" << endl;
BSONObj info;
conn.runCommand("istar", BSON("findandmodify" << "usr" << "query" << BSON("done" << BSON("$exists" << false) << "started" << BSON("$exists" << false)) << "sort" << BSON("submitted" << 1) << "update" << BSON("$set" << BSON("started" << Date_t(duration_cast<std::chrono::milliseconds>(system_clock::now().time_since_epoch()).count())))), info); // conn.findAndModify() is available since MongoDB C++ Driver legacy-1.0.0
const auto value = info["value"];
if (value.isNull())
{
// No incompleted jobs. Sleep for a while.
if (!sleeping) cout << local_time() << "Sleeping" << endl;
sleeping = true;
this_thread::sleep_for(chrono::seconds(10));
continue;
}
sleeping = false;
const auto job = value.Obj();
// Obtain job properties.
const auto _id = job["_id"].OID();
cout << local_time() << "Executing job " << _id.str() << endl;
const auto job_path = jobs_path / _id.str();
const auto format = job["format"].String();
const auto email = job["email"].String();
// Parse the user-supplied ligand.
OBMol obMol;
OBConversion obConversion;
obConversion.SetInFormat(format.c_str());
obConversion.ReadFile(&obMol, (job_path / ("ligand." + format)).string());
const auto num_atoms = obMol.NumAtoms();
// obMol.AddHydrogens(); // Adding hydrogens does not seem to affect SMARTS matching.
// Classify subset atoms.
array<vector<int>, num_subsets> subsets;
for (size_t k = 0; k < num_subsets; ++k)
{
auto& subset = subsets[k];
subset.reserve(num_atoms);
OBSmartsPattern smarts;
smarts.Init(SubsetSMARTS[k]);
smarts.Match(obMol);
for (const auto& map : smarts.GetMapList())
{
subset.push_back(map.front());
}
}
const auto& subset0 = subsets.front();
// Check user-provided ligand validity.
if (subset0.empty())
{
// Record job completion time stamp.
const auto millis_since_epoch = duration_cast<std::chrono::milliseconds>(system_clock::now().time_since_epoch()).count();
conn.update(collection, BSON("_id" << _id), BSON("$set" << BSON("done" << Date_t(millis_since_epoch))));
// Send error notification email.
cout << local_time() << "Sending an error notification email to " << email << endl;
MailMessage message;
message.setSender("usr <*****@*****.**>");
message.setSubject("Your usr job has failed");
message.setContent("Description: " + job["description"].String() + "\nSubmitted: " + to_simple_string(ptime(epoch, boost::posix_time::milliseconds(job["submitted"].Date().millis))) + " UTC\nFailed: " + to_simple_string(ptime(epoch, boost::posix_time::milliseconds(millis_since_epoch))) + " UTC\nReason: failed to parse the provided ligand.");
message.addRecipient(MailRecipient(MailRecipient::PRIMARY_RECIPIENT, email));
SMTPClientSession session("137.189.91.190");
session.login();
session.sendMessage(message);
session.close();
continue;
}
// Calculate the four reference points.
const auto n = subset0.size();
const auto v = 1.0 / n;
array<vector3, num_references> references{};
auto& ctd = references[0];
auto& cst = references[1];
auto& fct = references[2];
auto& ftf = references[3];
for (const auto i : subset0)
{
ctd += obMol.GetAtom(i)->GetVector();
}
ctd *= v;
double cst_dist = numeric_limits<double>::max();
double fct_dist = numeric_limits<double>::lowest();
double ftf_dist = numeric_limits<double>::lowest();
for (const auto i : subset0)
{
const auto& a = obMol.GetAtom(i)->GetVector();
const auto this_dist = a.distSq(ctd);
if (this_dist < cst_dist)
{
cst = a;
cst_dist = this_dist;
}
if (this_dist > fct_dist)
{
fct = a;
fct_dist = this_dist;
}
}
for (const auto i : subset0)
{
const auto& a = obMol.GetAtom(i)->GetVector();
const auto this_dist = a.distSq(fct);
if (this_dist > ftf_dist)
{
ftf = a;
ftf_dist = this_dist;
}
}
// Precalculate the distances between each atom and each reference point.
array<vector<double>, num_references> dista;
for (size_t k = 0; k < num_references; ++k)
{
const auto& reference = references[k];
auto& dists = dista[k];
dists.resize(1 + num_atoms); // OpenBabel atom index starts from 1. dists[0] is dummy.
for (size_t i = 0; i < n; ++i)
{
dists[subset0[i]] = sqrt(obMol.GetAtom(subset0[i])->GetVector().distSq(reference));
}
}
// Calculate USR and USRCAT features of the input ligand.
size_t qo = 0;
for (const auto& subset : subsets)
{
const auto n = subset.size();
for (size_t k = 0; k < num_references; ++k)
{
const auto& distp = dista[k];
vector<double> dists(n);
for (size_t i = 0; i < n; ++i)
{
dists[i] = distp[subset[i]];
}
array<double, 3> m{};
if (n > 2)
{
const auto v = 1.0 / n;
for (size_t i = 0; i < n; ++i)
{
const auto d = dists[i];
m[0] += d;
}
m[0] *= v;
for (size_t i = 0; i < n; ++i)
{
const auto d = dists[i] - m[0];
m[1] += d * d;
}
m[1] = sqrt(m[1] * v);
for (size_t i = 0; i < n; ++i)
{
const auto d = dists[i] - m[0];
m[2] += d * d * d;
}
m[2] = cbrt(m[2] * v);
}
else if (n == 2)
{
m[0] = 0.5 * (dists[0] + dists[1]);
m[1] = 0.5 * fabs(dists[0] - dists[1]);
}
else if (n == 1)
{
m[0] = dists[0];
}
#pragma unroll
for (const auto e : m)
{
q[qo++] = e;
}
}
}
assert(qo == qn.back());
// Compute USR and USRCAT scores.
usrcat_bin.seekg(0);
for (size_t k = 0; k < num_ligands; ++k)
{
usrcat_bin.read(reinterpret_cast<char*>(l.data()), sizeof(l));
double s = 0;
#pragma unroll
for (size_t i = 0, u = 0; u < num_usrs; ++u)
{
#pragma unroll
for (const auto qnu = qn[u]; i < qnu; ++i)
{
s += fabs(q[i] - l[i]);
}
scores[u][k] = s;
}
}
assert(usrcat_bin.tellg() == sizeof(l) * num_ligands);
// Sort ligands by USRCAT score and then by USR score and then by ZINC ID.
iota(scase.begin(), scase.end(), 0);
sort(scase.begin(), scase.end(), [&](const size_t val0, const size_t val1)
{
const auto u1score0 = u1scores[val0];
const auto u1score1 = u1scores[val1];
if (u1score0 == u1score1)
{
const auto u0score0 = u0scores[val0];
const auto u0score1 = u0scores[val1];
if (u0score0 == u0score1)
{
return zincids[val0] < zincids[val1];
}
return u0score0 < u0score1;
}
return u1score0 < u1score1;
});
// Write results.
filtering_ostream log_csv_gz;
log_csv_gz.push(gzip_compressor());
log_csv_gz.push(file_sink((job_path / "log.csv.gz").string()));
log_csv_gz.setf(ios::fixed, ios::floatfield);
log_csv_gz << "ZINC ID,USR score,USRCAT score\n" << setprecision(8);
filtering_ostream ligands_pdbqt_gz;
ligands_pdbqt_gz.push(gzip_compressor());
ligands_pdbqt_gz.push(file_sink((job_path / "ligands.pdbqt.gz").string()));
ligands_pdbqt_gz.setf(ios::fixed, ios::floatfield);
for (size_t t = 0; t < 10000; ++t)
{
const size_t k = scase[t];
const auto zincid = zincids[k].substr(0, 8); // Take another substr() to get rid of the trailing newline.
const auto u0score = 1 / (1 + scores[0][k] * qv[0]);
const auto u1score = 1 / (1 + scores[1][k] * qv[1]);
log_csv_gz << zincid << ',' << u0score << ',' << u1score << '\n';
// Only write conformations of the top ligands to ligands.pdbqt.gz.
if (t >= 1000) continue;
const auto zfp = zfproperties[k];
const auto zip = ziproperties[k];
ligands_pdbqt_gz
<< "MODEL " << '\n'
<< "REMARK 911 " << zincid
<< setprecision(3)
<< ' ' << setw(8) << zfp[0]
<< ' ' << setw(8) << zfp[1]
<< ' ' << setw(8) << zfp[2]
<< ' ' << setw(8) << zfp[3]
<< ' ' << setw(3) << zip[0]
<< ' ' << setw(3) << zip[1]
<< ' ' << setw(3) << zip[2]
<< ' ' << setw(3) << zip[3]
<< ' ' << setw(3) << zip[4]
<< '\n'
<< "REMARK 912 " << smileses[k] // A newline is already included in smileses[k].
<< "REMARK 913 " << suppliers[k] // A newline is already included in suppliers[k].
<< setprecision(8)
<< "REMARK 951 USR SCORE: " << setw(10) << u0score << '\n'
<< "REMARK 952 USRCAT SCORE: " << setw(10) << u1score << '\n'
;
const auto lig = ligands[k];
ligands_pdbqt_gz.write(lig.data(), lig.size());
ligands_pdbqt_gz << "ENDMDL\n";
}
// Update progress.
cout << local_time() << "Setting done time" << endl;
const auto millis_since_epoch = duration_cast<std::chrono::milliseconds>(system_clock::now().time_since_epoch()).count();
conn.update(collection, BSON("_id" << _id), BSON("$set" << BSON("done" << Date_t(millis_since_epoch))));
// Send completion notification email.
cout << local_time() << "Sending a completion notification email to " << email << endl;
MailMessage message;
message.setSender("istar <*****@*****.**>");
message.setSubject("Your usr job has completed");
message.setContent("Description: " + job["description"].String() + "\nSubmitted: " + to_simple_string(ptime(epoch, boost::posix_time::milliseconds(job["submitted"].Date().millis))) + " UTC\nCompleted: " + to_simple_string(ptime(epoch, boost::posix_time::milliseconds(millis_since_epoch))) + " UTC\nResult: http://istar.cse.cuhk.edu.hk/usr/iview/?" + _id.str());
message.addRecipient(MailRecipient(MailRecipient::PRIMARY_RECIPIENT, email));
SMTPClientSession session("137.189.91.190");
session.login();
session.sendMessage(message);
session.close();
}
}
int main(int argc, char* argv[])
{
// Check the required number of command line arguments.
if (argc != 5)
{
cout << "usr host user pwd jobs_path" << endl;
return 0;
}
// Fetch command line arguments.
const auto host = argv[1];
const auto user = argv[2];
const auto pwd = argv[3];
const path jobs_path = argv[4];
DBClientConnection conn;
{
// Connect to host and authenticate user.
cout << local_time() << "Connecting to " << host << " and authenticating " << user << endl;
string errmsg;
if ((!conn.connect(host, errmsg)) || (!conn.auth("istar", user, pwd, errmsg)))
{
cerr << local_time() << errmsg << endl;
return 1;
}
}
// Initialize constants.
cout << local_time() << "Initializing" << endl;
const auto collection = "istar.usr2";
const size_t num_usrs = 2;
const array<string, 2> usr_names{{ "USR", "USRCAT" }};
constexpr array<size_t, num_usrs> qn{{ 12, 60 }};
constexpr array<double, num_usrs> qv{{ 1.0 / qn[0], 1.0 / qn[1] }};
const size_t num_refPoints = 4;
const size_t num_subsets = 5;
const array<string, num_subsets> SubsetSMARTS
{{
"[!#1]", // heavy
"[#6+0!$(*~[#7,#8,F]),SH0+0v2,s+0,S^3,Cl+0,Br+0,I+0]", // hydrophobic
"[a]", // aromatic
"[$([O,S;H1;v2]-[!$(*=[O,N,P,S])]),$([O,S;H0;v2]),$([O,S;-]),$([N&v3;H1,H2]-[!$(*=[O,N,P,S])]),$([N;v3;H0]),$([n,o,s;+0]),F]", // acceptor
"[N!H0v3,N!H0+v4,OH+0,SH+0,nH+0]", // donor
}};
const size_t num_hits = 100;
// Wrap SMARTS strings to RWMol objects.
array<unique_ptr<ROMol>, num_subsets> SubsetMols;
for (size_t k = 0; k < num_subsets; ++k)
{
SubsetMols[k].reset(reinterpret_cast<ROMol*>(SmartsToMol(SubsetSMARTS[k])));
}
// Read ZINC ID file.
const string_array<size_t> zincids("16/zincid.txt");
const auto num_ligands = zincids.size();
cout << local_time() << "Found " << num_ligands << " database molecules" << endl;
// Read SMILES file.
const string_array<size_t> smileses("16/smiles.txt");
assert(smileses.size() == num_ligands);
// Read supplier file.
const string_array<size_t> suppliers("16/supplier.txt");
assert(suppliers.size() == num_ligands);
// Read property files of floating point types and integer types.
const auto zfproperties = read<array<float, 4>>("16/zfprop.f32");
assert(zfproperties.size() == num_ligands);
const auto ziproperties = read<array<int16_t, 5>>("16/ziprop.i16");
assert(ziproperties.size() == num_ligands);
// Read cumulative number of conformers file.
const auto mconfss = read<size_t>("16/mconfs.u64");
const auto num_conformers = mconfss.back();
assert(mconfss.size() == num_ligands);
assert(num_conformers >= num_ligands);
cout << local_time() << "Found " << num_conformers << " database conformers" << endl;
// Read feature file.
const auto features = read<array<double, qn.back()>>("16/usrcat.f64");
assert(features.size() == num_conformers);
// Read ligand footer file and open ligand SDF file for seeking and reading.
stream_array<size_t> ligands("16/ligand.sdf");
assert(ligands.size() == num_conformers);
// Initialize variables.
array<vector<int>, num_subsets> subsets;
array<vector<double>, num_refPoints> dista;
alignas(32) array<double, qn.back()> q;
// Initialize vectors to store compounds' primary score and their corresponding conformer.
vector<double> scores(num_ligands); // Primary score of molecules.
vector<size_t> cnfids(num_ligands); // ID of conformer with the best primary score.
const auto compare = [&](const size_t val0, const size_t val1) // Sort by the primary score.
{
return scores[val0] < scores[val1];
};
// Initialize an io service pool and create worker threads for later use.
const size_t num_threads = thread::hardware_concurrency();
cout << local_time() << "Creating an io service pool of " << num_threads << " worker threads" << endl;
io_service_pool io(num_threads);
safe_counter<size_t> cnt;
// Initialize the number of chunks and the number of molecules per chunk.
const auto num_chunks = num_threads << 4;
const auto chunk_size = 1 + (num_ligands - 1) / num_chunks;
assert(chunk_size * num_chunks >= num_ligands);
assert(chunk_size >= num_hits);
cout << local_time() << "Using " << num_chunks << " chunks and a chunk size of " << chunk_size << endl;
vector<size_t> scase(num_ligands);
vector<size_t> zcase(num_hits * (num_chunks - 1) + min(num_hits, num_ligands - chunk_size * (num_chunks - 1))); // The last chunk might have fewer than num_hits records.
// Enter event loop.
cout << local_time() << "Entering event loop" << endl;
cout.setf(ios::fixed, ios::floatfield);
bool sleeping = false;
while (true)
{
// Fetch an incompleted job in a first-come-first-served manner.
if (!sleeping) cout << local_time() << "Fetching an incompleted job" << endl;
BSONObj info;
const auto started = milliseconds_since_epoch();
conn.runCommand("istar", BSON("findandmodify" << "usr2" << "query" << BSON("started" << BSON("$exists" << false)) << "sort" << BSON("submitted" << 1) << "update" << BSON("$set" << BSON("started" << started))), info); // conn.findAndModify() is available since MongoDB C++ Driver legacy-1.0.0
const auto value = info["value"];
if (value.isNull())
{
// No incompleted jobs. Sleep for a while.
if (!sleeping) cout << local_time() << "Sleeping" << endl;
sleeping = true;
this_thread::sleep_for(chrono::seconds(2));
continue;
}
sleeping = false;
const auto job = value.Obj();
// Obtain job properties.
const auto _id = job["_id"].OID();
cout << local_time() << "Executing job " << _id.str() << endl;
const auto job_path = jobs_path / _id.str();
const size_t usr0 = job["usr"].Int(); // Specify the primary sorting score. 0: USR; 1: USRCAT.
assert(usr0 == 0 || usr0 == 1);
const auto usr1 = usr0 ^ 1;
const auto qnu0 = qn[usr0];
const auto qnu1 = qn[usr1];
// Read and validate the user-supplied SDF file.
cout << local_time() << "Reading and validating the query file" << endl;
SDMolSupplier sup((job_path / "query.sdf").string(), true, false, true); // sanitize, removeHs, strictParsing
if (!sup.length() || !sup.atEnd())
{
const auto error = 1;
cout << local_time() << "Failed to parse the query file, error code = " << error << endl;
conn.update(collection, BSON("_id" << _id), BSON("$set" << BSON("completed" << milliseconds_since_epoch() << "error" << error)));
continue;
}
// Process each of the query molecules sequentially.
const auto num_queries = 1; // Restrict the number of query molecules to 1. Setting num_queries = sup.length() to execute any number of query molecules.
for (unsigned int query_number = 0; query_number < num_queries; ++query_number)
{
cout << local_time() << "Parsing query molecule " << query_number << endl;
const unique_ptr<ROMol> qry_ptr(sup.next()); // Calling next() may print "ERROR: Could not sanitize molecule on line XXXX" to stderr.
auto& qryMol = *qry_ptr;
// Get the number of atoms, including and excluding hydrogens.
const auto num_atoms = qryMol.getNumAtoms();
const auto num_heavy_atoms = qryMol.getNumHeavyAtoms();
assert(num_heavy_atoms);
cout << local_time() << "Found " << num_atoms << " atoms and " << num_heavy_atoms << " heavy atoms" << endl;
// Create an output directory.
cout << local_time() << "Creating output directory" << endl;
const auto output_dir = job_path / to_string(query_number);
create_directory(output_dir);
// Draw a SVG.
cout << local_time() << "Drawing a SVG" << endl;
{
const unique_ptr<ROMol> qrz_ptr(removeHs(qryMol));
auto& qrzMol = *qrz_ptr;
compute2DCoords(qrzMol);
boost::filesystem::ofstream ofs(output_dir / "query.svg");
ofs << DrawingToSVG(MolToDrawing(qrzMol));
}
// Calculate Morgan fingerprint.
cout << local_time() << "Calculating Morgan fingerprint" << endl;
const unique_ptr<SparseIntVect<uint32_t>> qryFp(getFingerprint(qryMol, 2));
// Classify atoms to pharmacophoric subsets.
cout << local_time() << "Classifying atoms into subsets" << endl;
for (size_t k = 0; k < num_subsets; ++k)
{
vector<vector<pair<int, int>>> matchVect;
SubstructMatch(qryMol, *SubsetMols[k], matchVect);
const auto num_matches = matchVect.size();
auto& subset = subsets[k];
subset.resize(num_matches);
for (size_t i = 0; i < num_matches; ++i)
{
subset[i] = matchVect[i].front().second;
}
cout << local_time() << "Found " << num_matches << " atoms for subset " << k << endl;
}
const auto& subset0 = subsets.front();
assert(subset0.size() == num_heavy_atoms);
// Calculate the four reference points.
cout << local_time() << "Calculating " << num_refPoints << " reference points" << endl;
const auto qryRefPoints = calcRefPoints(qryMol, subset0);
const Point3DConstPtrVect qryRefPointv
{{
&qryRefPoints[0],
&qryRefPoints[1],
&qryRefPoints[2],
&qryRefPoints[3],
}};
// Precalculate the distances of heavy atoms to the reference points, given that subsets[1 to 4] are subsets of subsets[0].
cout << local_time() << "Calculating " << num_heavy_atoms * num_refPoints << " pairwise distances" << endl;
const auto& qryCnf = qryMol.getConformer();
for (size_t k = 0; k < num_refPoints; ++k)
{
const auto& refPoint = qryRefPoints[k];
auto& distp = dista[k];
distp.resize(num_atoms);
for (size_t i = 0; i < num_heavy_atoms; ++i)
{
distp[subset0[i]] = sqrt(dist2(qryCnf.getAtomPos(subset0[i]), refPoint));
}
}
// Loop over pharmacophoric subsets and reference points.
cout << local_time() << "Calculating " << 3 * num_refPoints * num_subsets << " moments of USRCAT feature" << endl;
size_t qo = 0;
for (const auto& subset : subsets)
{
const auto n = subset.size();
for (size_t k = 0; k < num_refPoints; ++k)
{
// Load distances from precalculated ones.
const auto& distp = dista[k];
vector<double> dists(n);
for (size_t i = 0; i < n; ++i)
{
dists[i] = distp[subset[i]];
}
// Compute moments.
array<double, 3> m{};
if (n > 2)
{
const auto v = 1.0 / n;
for (size_t i = 0; i < n; ++i)
{
const auto d = dists[i];
m[0] += d;
}
m[0] *= v;
for (size_t i = 0; i < n; ++i)
{
const auto d = dists[i] - m[0];
m[1] += d * d;
}
m[1] = sqrt(m[1] * v);
for (size_t i = 0; i < n; ++i)
{
const auto d = dists[i] - m[0];
m[2] += d * d * d;
}
m[2] = cbrt(m[2] * v);
}
else if (n == 2)
{
m[0] = 0.5 * (dists[0] + dists[1]);
m[1] = 0.5 * fabs(dists[0] - dists[1]);
}
else if (n == 1)
{
m[0] = dists[0];
}
for (const auto e : m)
{
q[qo++] = e;
}
}
}
assert(qo == qn.back());
// Compute USR and USRCAT scores.
cout << local_time() << "Calculating " << num_ligands << " " << usr_names[usr0] << " scores" << endl;
scores.assign(scores.size(), numeric_limits<double>::max());
iota(scase.begin(), scase.end(), 0);
cnt.init(num_chunks);
for (size_t l = 0; l < num_chunks; ++l)
{
io.post([&,l]()
{
// Loop over molecules of the current chunk.
const auto chunk_beg = chunk_size * l;
const auto chunk_end = min(chunk_beg + chunk_size, num_ligands);
for (size_t k = chunk_beg; k < chunk_end; ++k)
{
// Loop over conformers of the current molecule and calculate their primary score.
auto& scorek = scores[k];
size_t j = k ? mconfss[k - 1] : 0;
for (const auto mconfs = mconfss[k]; j < mconfs; ++j)
{
const auto& d = features[j];
double s = 0;
for (size_t i = 0; i < qnu0; ++i)
{
s += abs(q[i] - d[i]);
if (s >= scorek) break;
}
if (s < scorek)
{
scorek = s;
cnfids[k] = j;
}
}
}
// Sort the scores of molecules of the current chunk.
sort(scase.begin() + chunk_beg, scase.begin() + chunk_end, compare);
// Copy the indexes of top hits of the current chunk to a global vector for final sorting.
copy_n(scase.begin() + chunk_beg, min(num_hits, chunk_end - chunk_beg), zcase.begin() + num_hits * l);
cnt.increment();
});
}
cnt.wait();
// Sort the top hits from chunks.
cout << local_time() << "Sorting " << zcase.size() << " hits by " << usr_names[usr0] << " score" << endl;
sort(zcase.begin(), zcase.end(), compare);
// Create output directory and write output files.
cout << local_time() << "Writing output files" << endl;
SDWriter hits_sdf((output_dir / "hits.sdf").string());
boost::filesystem::ofstream hits_csv(output_dir / "hits.csv");
hits_csv.setf(ios::fixed, ios::floatfield);
hits_csv << "ZINC ID,USR score,USRCAT score,2D Tanimoto score,Molecular weight (g/mol),Partition coefficient xlogP,Apolar desolvation (kcal/mol),Polar desolvation (kcal/mol),Hydrogen bond donors,Hydrogen bond acceptors,Polar surface area tPSA (Å^2),Net charge,Rotatable bonds,SMILES,Vendors and annotations\n";
for (size_t l = 0; l < num_hits; ++l)
{
// Obtain indexes to the hit molecule and the hit conformer.
const auto k = zcase[l];
const auto j = cnfids[k];
// Read SDF content of the hit conformer.
const auto lig = ligands[j];
// Construct a RDKit ROMol object.
istringstream iss(lig);
SDMolSupplier sup(&iss, false, true, false, true);
assert(sup.length() == 1);
assert(sup.atEnd());
const unique_ptr<ROMol> hit_ptr(sup.next());
auto& hitMol = *hit_ptr;
// Calculate Morgan fingerprint.
const unique_ptr<SparseIntVect<uint32_t>> hitFp(getFingerprint(hitMol, 2));
// Calculate Tanimoto similarity.
const auto ts = TanimotoSimilarity(*qryFp, *hitFp);
// Find heavy atoms.
vector<vector<pair<int, int>>> matchVect;
SubstructMatch(hitMol, *SubsetMols[0], matchVect);
const auto num_matches = matchVect.size();
assert(num_matches == hitMol.getNumHeavyAtoms());
vector<int> hitHeavyAtoms(num_matches);
for (size_t i = 0; i < num_matches; ++i)
{
hitHeavyAtoms[i] = matchVect[i].front().second;
assert(hitHeavyAtoms[i] == i); // hitHeavyAtoms can be constructed using iota(hitHeavyAtoms.begin(), hitHeavyAtoms.end(), 0); because for RDKit-generated SDF molecules, heavy atom are always the first few atoms.
}
// Calculate the four reference points.
const auto hitRefPoints = calcRefPoints(hitMol, hitHeavyAtoms);
const Point3DConstPtrVect hitRefPointv
{{
&hitRefPoints[0],
&hitRefPoints[1],
&hitRefPoints[2],
&hitRefPoints[3],
}};
// Calculate a 3D transform from the four reference points of the hit conformer to those of the query molecule.
Transform3D trans;
AlignPoints(qryRefPointv, hitRefPointv, trans);
// Apply the 3D transform to all atoms of the hit conformer.
auto& hitCnf = hitMol.getConformer();
transformConformer(hitCnf, trans);
// Write the aligned hit conformer.
hits_sdf.write(hitMol);
// Calculate the secondary score of the saved conformer, which has the best primary score.
const auto& d = features[j];
double s = 0;
for (size_t i = 0; i < qnu1; ++i)
{
s += abs(q[i] - d[i]);
}
const auto u0score = 1 / (1 + scores[k] * qv[usr0]); // Primary score of the current molecule.
const auto u1score = 1 / (1 + s * qv[usr1]); // Secondary score of the current molecule.
const auto zincid = zincids[k].substr(0, 8); // Take another substr() to get rid of the trailing newline.
const auto zfp = zfproperties[k];
const auto zip = ziproperties[k];
const auto smiles = smileses[k]; // A newline is already included in smileses[k].
const auto supplier = suppliers[k]; // A newline is already included in suppliers[k].
hits_csv
<< zincid
<< setprecision(8)
<< ',' << (usr1 ? u0score : u1score)
<< ',' << (usr1 ? u1score : u0score)
<< ',' << ts
<< setprecision(3)
<< ',' << zfp[0]
<< ',' << zfp[1]
<< ',' << zfp[2]
<< ',' << zfp[3]
<< ',' << zip[0]
<< ',' << zip[1]
<< ',' << zip[2]
<< ',' << zip[3]
<< ',' << zip[4]
<< ',' << smiles.substr(0, smiles.length() - 1) // Get rid of the trailing newline.
<< ',' << supplier.substr(0, supplier.length() - 1) // Get rid of the trailing newline.
<< '\n'
;
}
}
// Update job status.
cout << local_time() << "Setting completed time" << endl;
const auto completed = milliseconds_since_epoch();
conn.update(collection, BSON("_id" << _id), BSON("$set" << BSON("completed" << completed << "nqueries" << num_queries)));
// Calculate runtime in seconds and screening speed in million conformers per second.
const auto runtime = (completed - started) * 0.001;
const auto speed = num_conformers * 0.000001 * num_queries / runtime;
cout
<< local_time() << "Completed " << num_queries << " " << (num_queries == 1 ? "query" : "queries") << " in " << setprecision(3) << runtime << " seconds" << endl
<< local_time() << "Screening speed was " << setprecision(0) << speed << " M conformers per second" << endl
;
}
}
static bool find_files(std::string pathname, const std::string &filename, string_array &files, bool recursive)
{
if (!pathname.empty())
{
char c = pathname[pathname.size() - 1];
if ((c != ':') && (c != '\\') && (c != '/'))
pathname += "\\";
}
WIN32_FIND_DATAA find_data;
HANDLE findHandle = FindFirstFileA((pathname + filename).c_str(), &find_data);
if (findHandle == INVALID_HANDLE_VALUE)
return false;
do
{
const bool is_directory = (find_data.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) != 0;
const bool is_system = (find_data.dwFileAttributes & FILE_ATTRIBUTE_SYSTEM) != 0;
const bool is_hidden = (find_data.dwFileAttributes & FILE_ATTRIBUTE_HIDDEN) != 0;
std::string filename(find_data.cFileName);
if ((!is_directory) && (!is_system) && (!is_hidden))
files.push_back(pathname + filename);
} while (FindNextFileA(findHandle, &find_data));
FindClose(findHandle);
if (recursive)
{
string_array paths;
HANDLE findHandle = FindFirstFileA((pathname + "*").c_str(), &find_data);
if (findHandle == INVALID_HANDLE_VALUE)
return false;
do
{
const bool is_directory = (find_data.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY) != 0;
const bool is_system = (find_data.dwFileAttributes & FILE_ATTRIBUTE_SYSTEM) != 0;
const bool is_hidden = (find_data.dwFileAttributes & FILE_ATTRIBUTE_HIDDEN) != 0;
std::string filename(find_data.cFileName);
if ((is_directory) && (!is_hidden) && (!is_system))
paths.push_back(filename);
} while (FindNextFileA(findHandle, &find_data));
FindClose(findHandle);
for (uint i = 0; i < paths.size(); i++)
{
const std::string &path = paths[i];
if (path[0] == '.')
continue;
if (!find_files(pathname + path, filename, files, true))
return false;
}
}
return true;
}
int main_internal(string_array cmd_line, int num_helper_threads, ilzham &lzham_dll)
{
comp_options options;
options.m_max_helper_threads = num_helper_threads;
#ifdef _XBOX
options.m_dict_size_log2 = 21;
#endif
if (!cmd_line.size())
{
print_usage();
return simple_test(lzham_dll, options);
}
enum op_mode_t
{
OP_MODE_INVALID = -1,
OP_MODE_COMPRESS = 0,
OP_MODE_DECOMPRESS = 1,
OP_MODE_ALL = 2
};
op_mode_t op_mode = OP_MODE_INVALID;
for (int i = 0; i < (int)cmd_line.size(); i++)
{
const std::string &str = cmd_line[i];
if (str[0] == '-')
{
if (str.size() < 2)
{
print_error("Invalid option: %s\n", str.c_str());
return EXIT_FAILURE;
}
switch (tolower(str[1]))
{
case 'u':
{
options.m_unbuffered_decompression = true;
break;
}
case 'd':
{
int dict_size = atoi(str.c_str() + 2);
if ((dict_size < LZHAM_MIN_DICT_SIZE_LOG2) || (dict_size > LZHAMTEST_MAX_POSSIBLE_DICT_SIZE))
{
print_error("Invalid dictionary size: %s\n", str.c_str());
return EXIT_FAILURE;
}
options.m_dict_size_log2 = dict_size;
break;
}
case 'm':
{
int comp_level = atoi(str.c_str() + 2);
if ((comp_level < 0) || (comp_level > (int)LZHAM_COMP_LEVEL_UBER))
{
print_error("Invalid compression level: %s\n", str.c_str());
return EXIT_FAILURE;
}
options.m_comp_level = static_cast<lzham_compress_level>(comp_level);
break;
}
case 't':
{
int num_threads = atoi(str.c_str() + 2);
if ((num_threads < 0) || (num_threads > LZHAM_MAX_HELPER_THREADS))
{
print_error("Invalid number of helper threads: %s\n", str.c_str());
return EXIT_FAILURE;
}
options.m_max_helper_threads = num_threads;
break;
}
case 'c':
{
options.m_compute_adler32_during_decomp = false;
break;
}
case 'v':
{
options.m_verify_compressed_data = true;
break;
}
case 'r':
{
options.m_randomize_params = true;
break;
}
case 'p':
{
options.m_force_polar_codes = true;
break;
}
case 'x':
{
options.m_extreme_parsing = true;
break;
}
case 'e':
{
options.m_deterministic_parsing = true;
break;
}
case 's':
{
int seed = atoi(str.c_str() + 2);
srand(seed);
printf("Using random seed: %i\n", seed);
break;
}
default:
{
print_error("Invalid option: %s\n", str.c_str());
return EXIT_FAILURE;
}
}
cmd_line.erase(cmd_line.begin() + i);
i--;
continue;
}
if (str.size() != 1)
{
print_error("Invalid mode: %s\n", str.c_str());
return EXIT_FAILURE;
}
switch (tolower(str[0]))
{
case 'c':
{
op_mode = OP_MODE_COMPRESS;
break;
}
case 'd':
{
op_mode = OP_MODE_DECOMPRESS;
break;
}
case 'a':
{
op_mode = OP_MODE_ALL;
break;
}
default:
{
print_error("Invalid mode: %s\n", str.c_str());
return EXIT_FAILURE;
}
}
cmd_line.erase(cmd_line.begin() + i);
break;
}
if (op_mode == OP_MODE_INVALID)
{
print_error("No mode specified!\n");
print_usage();
return EXIT_FAILURE;
}
printf("Using options:\n");
options.print();
printf("\n");
int exit_status = EXIT_FAILURE;
switch (op_mode)
{
case OP_MODE_COMPRESS:
{
if (cmd_line.size() < 2)
{
print_error("Must specify input and output filenames!\n");
return EXIT_FAILURE;
}
else if (cmd_line.size() > 2)
{
print_error("Too many filenames!\n");
return EXIT_FAILURE;
}
const std::string &src_file = cmd_line[0];
const std::string &cmp_file = cmd_line[1];
bool comp_result = compress_streaming(lzham_dll, src_file.c_str(), cmp_file.c_str(), options);
if (comp_result)
exit_status = EXIT_SUCCESS;
if ((comp_result) && (options.m_verify_compressed_data))
{
char decomp_file[256];
#ifdef _XBOX
sprintf(decomp_file, "e:\\__decomp_temp_%u__.tmp", (uint)GetTickCount());
#else
sprintf(decomp_file, "__decomp_temp_%u__.tmp", (uint)timer::get_ms());
#endif
if (!decompress_file(lzham_dll, cmp_file.c_str(), decomp_file, options))
{
print_error("Failed decompressing file \"%s\" to \"%s\"\n", cmp_file.c_str(), decomp_file);
return EXIT_FAILURE;
}
printf("Comparing file \"%s\" to \"%s\"\n", decomp_file, src_file.c_str());
if (!compare_files(decomp_file, src_file.c_str()))
{
print_error("Failed comparing decompressed file data while compressing \"%s\" to \"%s\"\n", src_file.c_str(), cmp_file.c_str());
return EXIT_FAILURE;
}
else
{
printf("Decompressed file compared OK to original file.\n");
}
remove(decomp_file);
}
break;
}
case OP_MODE_DECOMPRESS:
{
if (cmd_line.size() < 2)
{
print_error("Must specify input and output filenames!\n");
return EXIT_FAILURE;
}
else if (cmd_line.size() > 2)
{
print_error("Too many filenames!\n");
return EXIT_FAILURE;
}
if (decompress_file(lzham_dll, cmd_line[0].c_str(), cmd_line[1].c_str(), options))
exit_status = EXIT_SUCCESS;
break;
}
case OP_MODE_ALL:
{
if (!cmd_line.size())
{
print_error("No directory specified!\n");
return EXIT_FAILURE;
}
else if (cmd_line.size() != 1)
{
print_error("Too many filenames!\n");
return EXIT_FAILURE;
}
if (test_recursive(lzham_dll, cmd_line[0].c_str(), options))
exit_status = EXIT_SUCCESS;
break;
}
default:
{
print_error("No mode specified!\n");
print_usage();
return EXIT_FAILURE;
}
}
return exit_status;
}
bool MyCmdLineParser::Parse(const string_array& args)
{
// first parse the input command line into entries, don't care if they are valid or not
vector<string_array*> entries;
string_array pureArgs;
string_array* sa = NULL;
int argc = args.size();
for ( int i = 1; i < argc; i++ )
{
// hack to ignore mac os x bundle argument
string arg = args[i];
#ifdef Q_CYGWIN_WIN
arg = MyUtils::CygwinPathProof(args[i].c_str()).toStdString();
#endif
if (arg.substr(0, 6) != "-psn_0" )
{
if ( arg[0] == '-' && arg.length() > 1
&& !IsNumber( arg[1] ) && arg[1] != '.' )
{
sa = new string_array;
sa->clear();
if ( arg.length() > 2 && arg[1] == '-' ) // long name
{
sa->push_back( arg.substr(2) );
}
else
{
sa->push_back( arg.substr(1) );
}
entries.push_back( sa );
}
else if ( sa )
{
sa->push_back( arg );
}
else
{
pureArgs.push_back( arg );
}
}
}
//
m_cmdLineEntries.clear();
CmdLineEntry e;
bool bSucceed = true;
string error_msg = "";
for ( size_t i = 0; i < entries.size(); i++ )
{
string_array strgs = *entries[i];
if ( !IsValid( strgs[0].c_str(), &e ) ) // && !IsValid( strgs[0].c_str() + 1, &e ) )
{
bSucceed = false;
error_msg += "Option '" + strgs[0] + "' not recognized.";
break;
}
if ( e.type == CMD_LINE_OPTION )
{
e.arguments.clear();
for ( size_t j = 1; j < strgs.size(); j++ )
{
if ( j <= (size_t)e.maxArguments )
{
e.arguments.push_back( strgs[j] );
}
else
{
pureArgs.push_back( strgs[j] );
}
}
if ( (int)e.arguments.size() < e.minArguments )
{
bSucceed = false;
// cout << e.arguments.size() << " " << e.minArguments << "\n";
error_msg += "Argument missing for option '" + strgs[0] + "'.";
}
}
else if ( e.type == CMD_LINE_SWITCH )
{
for ( size_t j = 1; j < strgs.size(); j++ )
{
pureArgs.push_back( strgs[j] );
}
}
m_cmdLineEntries.push_back( e );
}
// release buffers
for ( size_t i = 0; i < entries.size(); i++ )
{
delete entries[i];
}
entries.clear();
/*
if ( bSucceed && (int)pureArgs.size() > m_nNumberOfPureArguments )
{
bSucceed = false;
error_msg += "Option '" + pureArgs[0] + "' not recognized.";
}
*/
m_cmdLineFloatingArguments = pureArgs;
if ( !bSucceed )
{
PrintErrorMessage( error_msg );
}
else if ( Found( "h" ) || Found( "help" ) )
{
PrintHelp();
bSucceed = false;
}
return bSucceed;
}
