pairwise_alignment_algorithm.hpp

Go to the documentation of this file.

// -----------------------------------------------------------------------------------------------------
// Copyright (c) 2006-2022, Knut Reinert & Freie Universität Berlin
// Copyright (c) 2016-2022, Knut Reinert & MPI für molekulare Genetik
// This file may be used, modified and/or redistributed under the terms of the 3-clause BSD-License
// shipped with this file and also available at: https://github.com/seqan/seqan3/blob/master/LICENSE.md
// -----------------------------------------------------------------------------------------------------
 
#pragma once
 
#include <concepts>
#include <ranges>
 
#include <seqan3/alignment/pairwise/detail/type_traits.hpp>
#include <seqan3/core/detail/empty_type.hpp>
#include <seqan3/utility/container/aligned_allocator.hpp>
#include <seqan3/utility/detail/type_name_as_string.hpp>
#include <seqan3/utility/simd/views/to_simd.hpp>
#include <seqan3/utility/views/elements.hpp>
 
namespace seqan3::detail
{
 
template <typename alignment_configuration_t, typename... policies_t>
    requires is_type_specialisation_of_v<alignment_configuration_t, configuration>
class pairwise_alignment_algorithm : protected policies_t...
{
protected:
    using traits_type = alignment_configuration_traits<alignment_configuration_t>;
    using score_type = typename traits_type::score_type;
    using alignment_result_type = typename traits_type::alignment_result_type;
 
    static_assert(!std::same_as<alignment_result_type, empty_type>, "Alignment result type was not configured.");
 
public:
    pairwise_alignment_algorithm() = default;                                                 
    pairwise_alignment_algorithm(pairwise_alignment_algorithm const &) = default;             
    pairwise_alignment_algorithm(pairwise_alignment_algorithm &&) = default;                  
    pairwise_alignment_algorithm & operator=(pairwise_alignment_algorithm const &) = default; 
    pairwise_alignment_algorithm & operator=(pairwise_alignment_algorithm &&) = default;      
    ~pairwise_alignment_algorithm() = default;                                                
 
    pairwise_alignment_algorithm(alignment_configuration_t const & config) : policies_t(config)...
    {}
 
    template <indexed_sequence_pair_range indexed_sequence_pairs_t, typename callback_t>
        requires std::invocable<callback_t, alignment_result_type>
    void operator()(indexed_sequence_pairs_t && indexed_sequence_pairs, callback_t && callback)
    {
        using std::get;
 
        for (auto && [sequence_pair, idx] : indexed_sequence_pairs)
        {
            size_t const sequence1_size = std::ranges::distance(get<0>(sequence_pair));
            size_t const sequence2_size = std::ranges::distance(get<1>(sequence_pair));
 
            auto && [alignment_matrix, index_matrix] = this->acquire_matrices(sequence1_size, sequence2_size);
 
            compute_matrix(get<0>(sequence_pair), get<1>(sequence_pair), alignment_matrix, index_matrix);
            this->make_result_and_invoke(std::forward<decltype(sequence_pair)>(sequence_pair),
                                         std::move(idx),
                                         this->optimal_score,
                                         this->optimal_coordinate,
                                         alignment_matrix,
                                         callback);
        }
    }
 
    template <indexed_sequence_pair_range indexed_sequence_pairs_t, typename callback_t>
        requires traits_type::is_vectorised && std::invocable<callback_t, alignment_result_type>
    auto operator()(indexed_sequence_pairs_t && indexed_sequence_pairs, callback_t && callback)
    {
        using simd_collection_t = std::vector<score_type, aligned_allocator<score_type, alignof(score_type)>>;
        using original_score_t = typename traits_type::original_score_type;
 
        // Extract the batch of sequences for the first and the second sequence.
        auto seq1_collection = indexed_sequence_pairs | views::elements<0> | views::elements<0>;
        auto seq2_collection = indexed_sequence_pairs | views::elements<0> | views::elements<1>;
 
        this->initialise_tracker(seq1_collection, seq2_collection);
 
        // Convert batch of sequences to sequence of simd vectors.
        thread_local simd_collection_t simd_seq1_collection{};
        thread_local simd_collection_t simd_seq2_collection{};
 
        convert_batch_of_sequences_to_simd_vector(simd_seq1_collection,
                                                  seq1_collection,
                                                  this->scoring_scheme.padding_symbol);
        convert_batch_of_sequences_to_simd_vector(simd_seq2_collection,
                                                  seq2_collection,
                                                  this->scoring_scheme.padding_symbol);
 
        size_t const sequence1_size = std::ranges::distance(simd_seq1_collection);
        size_t const sequence2_size = std::ranges::distance(simd_seq2_collection);
 
        auto && [alignment_matrix, index_matrix] = this->acquire_matrices(sequence1_size, sequence2_size);
 
        compute_matrix(simd_seq1_collection, simd_seq2_collection, alignment_matrix, index_matrix);
 
        size_t index = 0;
        for (auto && [sequence_pair, idx] : indexed_sequence_pairs)
        {
            original_score_t score = this->optimal_score[index]
                                   - (this->padding_offsets[index] * this->scoring_scheme.padding_match_score());
            matrix_coordinate coordinate{row_index_type{size_t{this->optimal_coordinate.row[index]}},
                                         column_index_type{size_t{this->optimal_coordinate.col[index]}}};
            this->make_result_and_invoke(std::forward<decltype(sequence_pair)>(sequence_pair),
                                         std::move(idx),
                                         std::move(score),
                                         std::move(coordinate),
                                         alignment_matrix,
                                         callback);
            ++index;
        }
    }
 
protected:
    template <typename simd_sequence_t, std::ranges::forward_range sequence_collection_t, arithmetic padding_symbol_t>
        requires std::ranges::output_range<simd_sequence_t, score_type>
    void convert_batch_of_sequences_to_simd_vector(simd_sequence_t & simd_sequence,
                                                   sequence_collection_t & sequences,
                                                   padding_symbol_t const & padding_symbol)
    {
        assert(static_cast<size_t>(std::ranges::distance(sequences)) <= traits_type::alignments_per_vector);
 
        simd_sequence.clear();
        for (auto && simd_vector_chunk : sequences | views::to_simd<score_type>(padding_symbol))
            std::ranges::move(simd_vector_chunk, std::back_inserter(simd_sequence));
    }
 
    template <std::ranges::forward_range sequence1_t,
              std::ranges::forward_range sequence2_t,
              std::ranges::input_range alignment_matrix_t,
              std::ranges::input_range index_matrix_t>
        requires std::ranges::forward_range<std::ranges::range_reference_t<alignment_matrix_t>>
              && std::ranges::forward_range<std::ranges::range_reference_t<index_matrix_t>>
    void compute_matrix(sequence1_t && sequence1,
                        sequence2_t && sequence2,
                        alignment_matrix_t && alignment_matrix,
                        index_matrix_t && index_matrix)
    {
        // ---------------------------------------------------------------------
        // Initialisation phase: allocate memory and initialise first column.
        // ---------------------------------------------------------------------
 
        this->reset_optimum(); // Reset the tracker for the new alignment computation.
 
        auto alignment_matrix_it = alignment_matrix.begin();
        auto indexed_matrix_it = index_matrix.begin();
 
        initialise_column(*alignment_matrix_it, *indexed_matrix_it, sequence2);
 
        // ---------------------------------------------------------------------
        // Iteration phase: compute column-wise the alignment matrix.
        // ---------------------------------------------------------------------
 
        for (auto alphabet1 : sequence1)
            compute_column(*++alignment_matrix_it,
                           *++indexed_matrix_it,
                           this->scoring_scheme_profile_column(alphabet1),
                           sequence2);
 
        // ---------------------------------------------------------------------
        // Final phase: track score of last column
        // ---------------------------------------------------------------------
 
        auto && alignment_column = *alignment_matrix_it;
        auto && cell_index_column = *indexed_matrix_it;
 
        auto alignment_column_it = alignment_column.begin();
        auto cell_index_column_it = cell_index_column.begin();
 
        this->track_last_column_cell(*alignment_column_it, *cell_index_column_it);
 
        for ([[maybe_unused]] auto && unused : sequence2)
            this->track_last_column_cell(*++alignment_column_it, *++cell_index_column_it);
 
        this->track_final_cell(*alignment_column_it, *cell_index_column_it);
    }
 
    template <std::ranges::input_range alignment_column_t,
              std::ranges::input_range cell_index_column_t,
              std::ranges::input_range sequence2_t>
    void initialise_column(alignment_column_t && alignment_column,
                           cell_index_column_t && cell_index_column,
                           sequence2_t && sequence2)
    {
        // ---------------------------------------------------------------------
        // Initial phase: prepare column and initialise first cell
        // ---------------------------------------------------------------------
 
        auto first_column_it = alignment_column.begin();
        auto cell_index_column_it = cell_index_column.begin();
        *first_column_it = this->track_cell(this->initialise_origin_cell(), *cell_index_column_it);
 
        // ---------------------------------------------------------------------
        // Iteration phase: iterate over column and compute each cell
        // ---------------------------------------------------------------------
 
        for ([[maybe_unused]] auto const & unused : sequence2)
        {
            ++first_column_it;
            *first_column_it =
                this->track_cell(this->initialise_first_column_cell(*first_column_it), *++cell_index_column_it);
        }
 
        // ---------------------------------------------------------------------
        // Final phase: track last cell of initial column
        // ---------------------------------------------------------------------
 
        this->track_last_row_cell(*first_column_it, *cell_index_column_it);
    }
 
    template <std::ranges::input_range alignment_column_t,
              std::ranges::input_range cell_index_column_t,
              typename alphabet1_t,
              std::ranges::input_range sequence2_t>
        requires semialphabet<alphabet1_t> || simd_concept<alphabet1_t>
    void compute_column(alignment_column_t && alignment_column,
                        cell_index_column_t && cell_index_column,
                        alphabet1_t const & alphabet1,
                        sequence2_t && sequence2)
    {
        using score_type = typename traits_type::score_type;
 
        // ---------------------------------------------------------------------
        // Initial phase: prepare column and initialise first cell
        // ---------------------------------------------------------------------
 
        auto alignment_column_it = alignment_column.begin();
        auto cell_index_column_it = cell_index_column.begin();
 
        auto cell = *alignment_column_it;
        score_type diagonal = cell.best_score();
        *alignment_column_it = this->track_cell(this->initialise_first_row_cell(cell), *cell_index_column_it);
 
        // ---------------------------------------------------------------------
        // Iteration phase: iterate over column and compute each cell
        // ---------------------------------------------------------------------
 
        for (auto const & alphabet2 : sequence2)
        {
            auto cell = *++alignment_column_it;
            score_type next_diagonal = cell.best_score();
            *alignment_column_it = this->track_cell(
                this->compute_inner_cell(diagonal, cell, this->scoring_scheme.score(alphabet1, alphabet2)),
                *++cell_index_column_it);
            diagonal = next_diagonal;
        }
 
        // ---------------------------------------------------------------------
        // Final phase: track last cell
        // ---------------------------------------------------------------------
 
        this->track_last_row_cell(*alignment_column_it, *cell_index_column_it);
    }
};
 
} // namespace seqan3::detail