Drug research and development time prediction method based on coding quantization and similarity matching
By considering the various influencing factors of drug R&D projects based on the method of coding quantification and similarity matching, the problem of inaccurate prediction of drug R&D projects in the existing technology is solved, and more accurate and practical prediction results are achieved.
Patent Information
- Application Number
- CN202411951825.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-27
- Publication Date
- 2025-04-15
AI Technical Summary
The prior art is difficult to achieve accurate prediction of drug research and development project time based on the full consideration of various influencing factors, especially when ignoring the differences between different treatment fields and original research units.
Using a method based on encoding quantization and similarity matching, information encoding processing is performed on historical items and items to be predicted through preset encoding rules, encoding vectors are generated and stored in the project encoding database. Then, based on the coded vector of the items to be predicted, the historical items with the highest similarity are matched from the database, the average duration of each reference item is calculated, and a variety of influencing factors such as treatment areas and original research units are considered, and the prediction results are refined through screening conditions and mean duration deviations.
It realizes accurate prediction of drug R&D project time, improves the accuracy and practicality of prediction, can more comprehensively consider the complexity and uncertainty of drug R&D projects, and provides more accurate prediction results, helping drug R&D companies better plan and manage R&D projects.
Smart Images
Figure CN119831103A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of big data analysis and prediction, and in particular relates to a drug development time prediction method based on coding quantization and similarity matching. Background Art
[0002] In drug development, accurate prediction of project time is crucial for project management, resource allocation and decision making. However, the drug development process is complex and changeable, involving multiple development stages and multiple influencing factors, such as drug classification, drug type, company location, therapeutic area and original research unit. These factors make the time prediction of drug development projects extremely challenging.
[0003] In the prior art, the time prediction of drug development projects mainly relies on expert experience, historical data statistics and simple mathematical models. However, these methods have many limitations. Although expert experience has certain reference value, it is limited by personal experience and knowledge, and it is difficult to guarantee the accuracy and consistency of the prediction results. Although the historical data statistical method can provide certain data support, it often ignores the differences between different projects, resulting in overly general prediction results. Simple mathematical models may not be able to fully consider various influencing factors, resulting in a large deviation between the prediction results and the actual situation.
[0004] In particular, in the existing technology, the factors affecting the time of drug development projects are not considered comprehensively and deeply enough. Drug development projects in different therapeutic areas and original research units may have significant differences in development process and time. However, the existing technology often ignores these differences, resulting in the prediction results that cannot accurately reflect the actual situation.
[0005] Therefore, how to achieve accurate prediction of the time of drug development projects based on full consideration of various influencing factors has become an urgent problem to be solved. Summary of the invention
[0006] In view of the above-mentioned deficiencies in the prior art, the present invention provides a drug development time prediction method based on coding quantization and similarity matching, which can achieve accurate prediction of the time of drug development projects on the basis of fully considering various influencing factors.
[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions: The drug development time prediction method based on coding quantization and similarity matching includes the following steps: S1. Obtain historical items, perform information encoding processing according to preset encoding rules, obtain corresponding encoding vectors and store them in the item encoding database; S2. According to the same coding rules, information encoding is performed on the drug development project to be predicted to obtain a corresponding coding vector; S3, according to the coding vector of the drug development project to be predicted, matching N historical projects with the highest coding vector similarity from the project coding database as reference projects; S4. Based on the R&D stage to be predicted, calculate the average duration of each reference project in the corresponding R&D stage, and record it as the general average duration; S5. Determine the influencing factors of the drug development project to be predicted; for each influencing factor, formulate screening conditions for it, and screen out reference projects that meet the screening conditions from the reference projects obtained in S3 as factor screening projects, and then calculate the average duration of the factor screening projects in the development stage to be predicted, and compare it with the general average duration to obtain the average duration deviation of the influencing factor; S6. Calculate the predicted completion date of the drug development project to be predicted in the research and development stage to be predicted based on the general average duration obtained in S4 and the average duration deviation of each influencing factor obtained in S5.
[0008] Compared with the prior art, the present invention has the following beneficial effects: 1. Data standardization and efficient storage. The information encoding process of historical projects and projects to be predicted is carried out through preset encoding rules, which realizes the standardization and unified format of project data, and facilitates subsequent data processing and storage. The encoding vector is stored in the project encoding database, which improves the efficiency of data retrieval and matching and provides a solid foundation for subsequent similarity calculation.
[0009] 2. Accuracy of similarity matching. Using encoding vectors for similarity matching can accurately find the historical project that is most similar to the project to be predicted, providing a reliable reference for prediction. Compared with traditional text or keyword-based matching methods, encoding vector similarity matching can capture more complex project features and information, improving matching accuracy.
[0010] 3. Consider multiple influencing factors. This method not only considers the average R&D time of historical projects, but also introduces multiple influencing factors (such as therapeutic areas, original research units, etc.). By formulating screening conditions and calculating the average time deviation, the prediction results are further refined. This method of comprehensively considering multiple influencing factors makes the prediction results closer to the actual situation and improves the accuracy and practicality of the prediction.
[0011] 4. It can accurately predict the drug development time. On the basis of fully considering various influencing factors, this method calculates the general average duration and the average duration deviation to obtain the predicted completion date of the drug development project to be predicted in the predicted development stage. Compared with the existing prediction methods based on experience or statistical models, this technical solution can more accurately reflect the complexity and uncertainty of drug development projects and provide more accurate prediction results.
[0012] Compared with the existing technology, this technical solution has shown significant superiority in prediction accuracy, practicality and operability. It can more comprehensively consider the complexity and uncertainty of drug R&D projects, provide more accurate prediction results, and help drug R&D companies better plan and manage R&D projects.
[0013] In summary, this method can achieve accurate prediction of the time of drug development projects on the basis of fully considering various influencing factors, and provide strong decision-making support for drug development companies.
[0014] Preferably, in S1, the preset coding rules include uniqueness, comprehensive coverage of key information, scalability and readability.
[0015] This setting, 1. Uniqueness ensures that each drug development project has a unique identifier in the coding database, avoiding information confusion and misjudgment. This uniqueness enables the accurate finding of the historical project most similar to the project to be predicted in the subsequent similarity matching process, improving the matching accuracy and prediction accuracy.
[0016] 2. Coding rules require the coverage of key information of the project, such as drug type, therapeutic area, etc. The comprehensiveness of this information ensures the accuracy and reliability of the prediction results. The key information extracted through coding rules can fully reflect the characteristics and complexity of drug development projects, and provide rich and accurate data support for similarity matching and prediction.
[0017] 3. Scalability enables the coding rules to adapt to new features and requirements that may arise in future drug development projects. As the field of drug development continues to develop and change, new key information or features may be identified. The scalability of the coding rules ensures that this new information can be effectively incorporated into the coding system, thereby maintaining the timeliness and accuracy of the prediction method.
[0018] 4. Readability makes the encoding vector easy to understand and process for both humans and machines. In the similarity matching and prediction process, easy-to-read encoding vectors can reduce algorithm complexity and computational costs, and improve processing speed and efficiency. At the same time, easy-to-read encoding vectors also help drug research and development companies better understand the prediction results and make more informed decisions.
[0019] Preferably, the information encoding process includes: Obtain key information of the project; Encode each key information of the project according to a preset encoding rule to obtain the encoding of each key information; Combine the codes of each key information according to the preset coding structure sequence to form a complete code; The generated code is verified to ensure that it meets the requirements of the coding rules; and the verified code is used as the information coding processing result.
[0020] Such a setting, 1. By obtaining key information of the project (such as drug classification, drug type, company name, company location) and converting this information according to unified coding rules, standardization and consistency of information are achieved. This helps to ensure that information from different sources and formats can be understood and processed in a unified way.
[0021] 2. The encoding process converts complex information into codes, which greatly reduces the complexity and redundancy of the data and improves the processability of the data. This makes the subsequent data analysis, similarity matching and prediction processes more efficient and accurate.
[0022] Preferably, the key information of the project includes therapeutic area, indication, special review, project name, original research unit, highest stage, drug classification, drug type and company location; When encoding the key information of the project, the one-hot coding method is used to encode special review, project name, original research unit, highest stage, drug classification, drug type and company location.
[0023] Such a setting, 1. By incorporating key information such as therapeutic areas and indications into the coding scope, this method can more comprehensively reflect the characteristics and background of drug development projects. This information is crucial for assessing the complexity and potential risks of the project, thereby helping to improve the accuracy of the prediction.
[0024] 2. One-hot encoding is a commonly used classification data encoding method, which can map each category of the classification variable to a vector containing only 0 and 1. This method is very effective when dealing with classification data with a limited number of categories, such as special reviews, project names (although project names are usually not classification variables, they may refer to the unique identifier or classification label of the project here), original research units, highest stage, drug classification, drug type and company location. Through one-hot encoding, these classification information can be converted into numerical forms that can be processed by machine learning models, thereby facilitating subsequent similarity matching and predictive analysis.
[0025] Preferably, when encoding the key information of the project, the process of encoding the therapeutic areas and indications includes: First, determine the number of categories of this key information; Then, all categories of the key information are arranged in a preset order to obtain a sequence of the key information to be filled; Then, the coding values of each category of the key information are calculated according to the preset method; Finally, fill the code value of each category into the corresponding position of the sequence to be filled to obtain the code of the key information.
[0026] Such a setting can achieve standardization and normalization. By determining the number of categories of key information and arranging all categories in a preset order, the consistency and standardization of the coding process can be ensured. This helps to eliminate the subjectivity and differences in the coding process between different people or at different time points, thereby improving the accuracy and reliability of coding.
[0027] 2. A preset calculation method is used to determine the code value for each category. This is usually an efficient and systematic method that can quickly generate the code for each category. Compared with manual coding or coding methods that rely on complex algorithms, this method is easier to implement and maintain, thereby improving the efficiency of coding.
[0028] 3. Using this method, the encoding vector of the project information obtained by combining the key information can effectively ensure the intuitiveness (i.e. which categories of specific key information are the same or similar) and accuracy of the similarity comparison when performing similarity comparison.
[0029] Preferably, the TF-IDF encoding formula is used to calculate the encoding value of each category.
[0030] Such a setting can, 1. increase the discrimination of the encoding. The TF-IDF algorithm combines the two factors of term frequency (TF) and inverse document frequency (IDF), and can evaluate the importance of a word in a document. In drug development projects, key information such as therapeutic areas and indications are regarded as "words" and projects are regarded as "documents". Through TF-IDF encoding, the importance of each category (that is, the different values of each key information) in the overall project collection can be calculated. This method allows categories that appear frequently in the project but rarely appear in other projects to obtain higher encoding values, thereby enhancing the discrimination of the encoding. This helps to more accurately identify historical projects similar to the project to be predicted in the subsequent similarity matching process.
[0031] 2. It can improve the accuracy of prediction. Since TF-IDF coding can more accurately reflect the importance of each category, the use of TF-IDF coding in the drug development time prediction method based on coding quantification and similarity matching can improve the accuracy of prediction. By matching historical projects with high similarity to the project to be predicted, and predicting the development time of the project to be predicted based on the development time data of these projects, more reliable prediction results can be obtained.
[0032] Preferably, in S3, a cosine similarity algorithm is used to calculate the similarity of the encoding vectors of the drug development project to be predicted and the historical projects.
[0033] In this setting, 1. The cosine similarity algorithm measures the similarity of two vectors by calculating the cosine of the angle between them in multidimensional space. This method can accurately capture the direction and correlation between vectors, even if their absolute values are different. In the context of drug development project encoding vectors, this means that the algorithm can accurately identify the historical project that is closest to the project to be predicted in terms of key information.
[0034] 2. The encoding vectors of drug development projects usually contain information of multiple dimensions (such as drug classification, drug type, company name, company location, etc.), which constitute a high-dimensional space. The cosine similarity algorithm performs well in processing high-dimensional data and can quickly and accurately calculate the similarity between vectors, which is particularly important for matching and analyzing large-scale data sets.
[0035] 3. Strong adaptability. The cosine similarity algorithm does not depend on the specific values of the vectors, but focuses on the direction and relative size between them. This makes the algorithm highly adaptable to different types of coding vectors (such as binary vectors, real vectors, etc.). In the coding vectors of drug development projects, no matter what encoding method or value range is used, the cosine similarity algorithm can effectively calculate the similarity.
[0036] Preferably, in S5, the process of determining the influencing factors of the drug development project to be predicted includes: based on historical project data, quantitatively analyzing various factors of the drug development project to determine their degree of influence on the development time; taking the M factors with the highest degree of influence as influencing factors, and determining the weight value of each influencing factor according to the degree of influence.
[0037] Such a setting can, 1. Improve the accuracy of prediction. By quantitatively analyzing the impact of various factors on R&D time, it is possible to more accurately identify those factors that have a significant impact on R&D time. This helps to consider these factors more accurately in the subsequent prediction process, thereby improving the accuracy of the prediction. The determination of the weight value allows the importance of each influencing factor in the prediction process to be quantified. This helps decision makers to more scientifically evaluate the impact of different factors on R&D time, thereby making more informed and reasonable decisions.
[0038] 2. It can improve project management. By identifying key influencing factors and determining their weight values, we can have a more comprehensive understanding of the characteristics and needs of R&D projects. This helps improve project management, including project planning, risk identification and response, and progress monitoring.
[0039] Preferably, in S6, the predicted completion date of the drug development project to be predicted in the research and development stage to be predicted is calculated by the following formula: Predicted completion time = current date + general average duration + average duration difference of each influencing factor × weight value.
[0040] With this setting, the formula takes into account multiple influencing factors and their weight values, which reflects the different degrees of influence of different factors on R&D time. By quantifying these factors, the completion of R&D projects can be predicted more accurately, improving the accuracy and reliability of the prediction. The formula combines the current date, the general average duration, the average duration difference of each influencing factor, and the weight value to calculate the specific predicted completion time. This provides a clear time reference for project managers and decision makers, helping them better plan project schedules and resource allocation.
[0041] Preferably, the influencing factors include therapeutic areas and original research units.
[0042] Such a setting can improve the accuracy of predictions. Therapeutic areas and original research units are two key factors in drug development projects, which have a direct and significant impact on the development time. For example, drug development in different therapeutic areas may face different technical challenges and regulatory requirements, and the research strength, experience and resources of the original research unit will also directly affect the development progress. Incorporating these two factors into the prediction model can more accurately reflect the actual situation of the project, thereby improving the accuracy of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to make the purpose, technical solution and advantages of the invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings, in which: Figure 1 Flowchart of this method. DETAILED DESCRIPTION
[0044] The following is a further detailed description through specific implementation methods: Embodiment 1 like Figure 1 As shown, this embodiment discloses a method for predicting drug development time based on coding quantization and similarity matching, comprising the following steps: S1. Obtain historical projects, and perform information encoding processing according to preset encoding rules to obtain corresponding encoding vectors and store them in the project encoding database.
[0045] Among them, the preset coding rules include uniqueness, comprehensive coverage of key information, scalability and readability.
[0046] In this way, uniqueness ensures that each drug development project has a unique identifier in the coding database, avoiding information confusion and misjudgment. This uniqueness enables the accurate finding of historical projects that are most similar to the project to be predicted in the subsequent similarity matching process, improving the accuracy of matching and the accuracy of prediction. In addition, the coding rules require the coverage of key information of the project, such as drug type, therapeutic area, etc. The comprehensiveness of this information ensures the accuracy and reliability of the prediction results. The key information extracted through the coding rules can fully reflect the characteristics and complexity of drug development projects, providing rich and accurate data support for similarity matching and prediction.
[0047] Scalability enables the coding rules to adapt to new features and requirements that may arise in future drug development projects. As the field of drug development continues to grow and change, new key information or features may be identified. The scalability of the coding rules ensures that this new information can be effectively incorporated into the coding system, thereby maintaining the timeliness and accuracy of the prediction method. Readability makes the coding vector easy to understand and process for both humans and machines. In the process of similarity matching and prediction, easy-to-read coding vectors can reduce algorithm complexity and computational costs, and improve processing speed and efficiency. At the same time, easy-to-read coding vectors also help drug development companies better understand the prediction results and make more informed decisions.
[0048] In specific implementation, the information encoding process includes: Obtain key information of the project, including therapeutic area, indication, special review, project name, original research unit, highest stage, drug classification, drug type and company location; Encode each key information of the project according to a preset encoding rule to obtain the encoding of each key information; Combine the codes of each key information according to the preset coding structure sequence to form a complete code; The generated code is verified to ensure that it meets the requirements of the coding rules; and the verified code is used as the information coding processing result.
[0049] In this way, by obtaining key information of the project (such as drug classification, drug type, company name, company location) and converting this information according to unified coding rules, standardization and consistency of information are achieved. This helps to ensure that information from different sources and formats can be understood and processed in a unified way. In addition, the coding process converts complex information into concise codes, which greatly reduces the complexity and redundancy of the data and improves the processability of the data. This makes the subsequent data analysis, similarity matching and prediction processes more efficient and accurate.
[0050] During specific implementation, when encoding the key information of the project, the one-hot coding method is used to encode the special review, project name, original research unit, highest stage, drug classification, drug type and company location.
[0051] One-hot encoding is a commonly used categorical data encoding method that maps each category of a categorical variable to a vector containing only 0 and 1. This method is very effective when dealing with categorical data with a limited number of categories, such as special reviews, project names (although project names are usually not categorical variables, they may refer to the unique identifier or classification label of the project here), original research units, highest stage, drug classification, drug type, and company location. Through one-hot encoding, this categorical information can be converted into a numerical form that can be processed by machine learning models, thereby facilitating subsequent similarity matching and predictive analysis.
[0052] When coding the key information of the project, the process of coding the therapeutic areas and indications includes: First, determine the number of categories of this key information; Then, all categories of the key information are arranged in a preset order to obtain a sequence of the key information to be filled; Then, use the TF-IDF coding formula to calculate the coding values of each category of the key information; Finally, fill the code value of each category into the corresponding position of the sequence to be filled to obtain the code of the key information.
[0053] In this way, standardization and normalization can be achieved. By determining the number of categories for key information and arranging all categories in a preset order, consistency and standardization of the coding process can be ensured. This helps to eliminate subjectivity and differences in the coding process between different people or at different time points, thereby improving the accuracy and reliability of coding. Preset calculation methods are used to determine the coding values for each category, which is usually an efficient and systematic method to quickly generate codes for each category. Compared with manual coding or coding methods that rely on complex algorithms, this method is easier to implement and maintain, thereby improving the efficiency of coding.
[0054] In addition, the TF-IDF algorithm combines the two factors of term frequency (TF) and inverse document frequency (IDF), which can evaluate the importance of a word in a document. In drug development projects, key information such as therapeutic areas and indications are regarded as "words" and projects are regarded as "documents". Through TF-IDF encoding, the importance of each category (that is, the different values of each key information) in the overall project collection can be calculated. This method allows categories that appear frequently in projects but rarely appear in other projects to obtain higher coding values, thereby enhancing the distinctiveness of the coding. This helps to more accurately identify historical projects similar to the project to be predicted in the subsequent similarity matching process. Since TF-IDF encoding can more accurately reflect the importance of each category, the use of TF-IDF encoding in the drug development time prediction method based on coding quantification and similarity matching can improve the accuracy of prediction.
[0055] By using this method, the encoding vectors of the project information obtained by combining the key information can effectively ensure the intuitiveness (i.e. which categories of the specific key information are the same or similar) and accuracy of the similarity comparison when performing similarity comparison.
[0056] S2. According to the same coding rules, information encoding is performed on the drug development project to be predicted to obtain the corresponding coding vector.
[0057] S3. According to the coding vector of the drug development project to be predicted, N historical projects with the highest coding vector similarity are matched from the project coding database as reference projects. The specific value of N can be set by those skilled in the art according to the scale of the project coding database and the accuracy requirement of the prediction during specific use, and will not be described in detail here.
[0058] In specific implementation, the cosine similarity algorithm is used to calculate the similarity of the coding vectors between the drug development project to be predicted and the historical projects.
[0059] Thus, the cosine similarity algorithm measures the similarity of two vectors by calculating the cosine value of the angle between them in a multidimensional space. This method can accurately capture the direction and correlation between vectors, even if their absolute values are different. In the context of the coding vectors of drug development projects, this means that the algorithm can accurately identify the historical project that is closest to the project to be predicted in terms of key information. In addition, the coding vectors of drug development projects usually contain information in multiple dimensions (such as drug classification, drug type, company name, company location, etc.), which constitute a high-dimensional space. The cosine similarity algorithm performs well in processing high-dimensional data and can quickly and accurately calculate the similarity between vectors, which is particularly important for matching and analyzing large-scale data sets. And it is highly adaptable. The cosine similarity algorithm does not rely on the specific values of the vectors, but focuses on the direction and relative size between them. This makes the algorithm highly adaptable to different types of coding vectors (such as binary vectors, real number vectors, etc.). In the coding vectors of drug development projects, the cosine similarity algorithm can effectively calculate the similarity regardless of the encoding method or value range used.
[0060] S4. Based on the R&D stage to be predicted, calculate the average duration of each reference project in the corresponding R&D stage and record it as the universal average duration.
[0061] S5. Determine the influencing factors of the drug development project to be predicted; for each influencing factor, formulate screening conditions, and select reference projects that meet the screening conditions from the reference projects obtained in S3 as factor screening projects, and then calculate the average duration of the factor screening projects in the research and development stage to be predicted, and compare it with the general average duration to obtain the average duration deviation of the influencing factor.
[0062] In specific implementation, the process of determining the influencing factors of the drug development project to be predicted includes: based on historical project data, quantitatively analyzing various factors of the drug development project to determine their degree of influence on the development time; taking the M factors with the highest degree of influence as influencing factors, and determining the weight value of each influencing factor according to the degree of influence.
[0063] In this way, the accuracy of the forecast can be improved. By quantitatively analyzing the impact of various factors on R&D time, those factors that have a significant impact on R&D time can be more accurately identified. This helps to consider these factors more accurately in the subsequent forecasting process, thereby improving the accuracy of the forecast. The determination of weight values allows the importance of each influencing factor in the forecasting process to be quantified. This helps decision makers to more scientifically evaluate the impact of different factors on R&D time, so as to make more informed and reasonable decisions. It can also improve the level of project management. By identifying key influencing factors and determining their weight values, the characteristics and needs of R&D projects can be more comprehensively understood. This helps to improve the level of project management, including project planning, risk identification and response, and progress monitoring.
[0064] In this embodiment, the influencing factors include therapeutic areas and original research units. The original research unit refers to the unit that originally developed the drug or similar drugs (which may be on the market or in the research and development stage). In the pharmaceutical industry, the original research unit often refers to those companies that initially developed a new drug or have innovative drug technology. Such a setting can improve the accuracy of the prediction. The therapeutic field and the original research unit are two key factors in the drug research and development project, and they have a direct and significant impact on the research and development time. For example, drug research and development in different therapeutic areas may face different technical challenges and regulatory requirements, and the research strength, experience and resources of the original research unit will also directly affect the progress of research and development. Incorporating these two factors into the prediction model can more accurately reflect the actual situation of the project, thereby improving the accuracy of the prediction.
[0065] S6. Calculate the predicted completion date of the drug development project to be predicted in the research and development stage to be predicted based on the general average duration obtained in S4 and the average duration deviation of each influencing factor obtained in S5.
[0066] In specific implementation, the following formula is used to calculate the predicted completion date of the drug development project to be predicted in the research and development stage to be predicted: Predicted completion time = current date + general average duration + average duration difference of each influencing factor × weight value.
[0067] With this setting, the formula takes into account multiple influencing factors and their weight values, which reflects the different degrees of influence of different factors on R&D time. By quantifying these factors, the completion of R&D projects can be predicted more accurately, improving the accuracy and reliability of the prediction. The formula combines the current date, the general average duration, the average duration difference of each influencing factor, and the weight value to calculate the specific predicted completion time. This provides a clear time reference for project managers and decision makers, helping them better plan project schedules and resource allocation.
[0068] This method uses preset coding rules to encode information of historical projects and projects to be predicted, realizes the standardization and unified format of project data, and facilitates subsequent data processing and storage. The coding vector is stored in the project coding database, which improves the efficiency of data retrieval and matching, and provides a solid foundation for subsequent similarity calculation. In addition, using coding vectors for similarity matching can accurately find the historical projects most similar to the project to be predicted, providing a reliable reference for prediction. Compared with traditional text-based or keyword-based matching methods, coding vector similarity matching can capture more complex project features and information, and improve the accuracy of matching. In addition, this method not only considers the average R&D time of historical projects, but also introduces a variety of influencing factors (such as therapeutic fields, original research units, etc.), and further refines the prediction results by formulating screening conditions and calculating the average time deviation. This method that comprehensively considers multiple influencing factors makes the prediction results closer to the actual situation and improves the accuracy and practicality of the prediction.
[0069] This method can accurately predict the drug development time. On the basis of fully considering various influencing factors, the technical solution calculates the general average duration and the average duration deviation to obtain the predicted completion date of the drug development project to be predicted in the research and development stage to be predicted. Compared with the existing prediction methods based on experience or statistical models, this technical solution can more accurately reflect the complexity and uncertainty of drug development projects and provide more accurate prediction results. Compared with the prior art, this technical solution shows significant superiority in prediction accuracy, practicality and operability. It can more comprehensively consider the complexity and uncertainty of drug development projects, provide more accurate prediction results, and help drug development companies to better plan and manage research and development projects. In summary, this method can achieve accurate prediction of the time of drug development projects on the basis of fully considering various influencing factors, providing strong decision-making support for drug development companies.
[0070] Embodiment 2 In order to better understand the encoding of key information of the project in this method, the following explanation is given.
[0071] In the drug development time prediction method based on coding quantification and similarity matching, encoding the key information of the project is a crucial step. The following is the specific process of encoding the key information of the project, including the methods of using One-hot encoding and TF-IDF encoding.
[0072] 1. Coding for “special review, project name, original research unit, highest stage, drug classification, drug type and company location”.
[0073] For categorical data with only one category number (special review, project name, original research unit, highest stage, drug classification, drug type and company location), one-hot encoding method was used for encoding.
[0074] Special review: Assume that there are three categories of special review: fast track, breakthrough therapy, and orphan drug. For each project, if it belongs to a certain category, the position of the category is coded as 1, and the rest are 0. For example, if project A is a fast track, its code is [1,0,0].
[0075] Project name: Here, "project name" refers to the unique identifier or classification label of the project. Assuming there are five projects, each project name can be encoded as a five-bit binary number, where only one bit is 1 and the rest are 0. For example, the encoding of project B is [0,1,0,0,0].
[0076] Original research unit: Assume there are three original research units: Unit A, Unit B, and Unit C. For each project, if it is developed by original research unit A, it is coded as [1,0,0], if it is developed by original research unit B, it is coded as [0,1,0], and if it is developed by original research unit C, it is coded as [0,0,1].
[0077] Highest stage: Assume that there are four highest stages of drug development: preclinical, phase I clinical, phase II clinical, phase III clinical and beyond. For each project, if its highest stage is a certain category, the position of that category is coded as 1, and the rest are 0. For example, the highest stage of project C is phase II clinical, so its code is [0,0,1,0].
[0078] Drug classification: Assuming there are three drug classifications: chemical drugs, biological drugs, and traditional Chinese medicine, the coding method is similar to special review.
[0079] Drug type: Assume there are two types of drugs: innovative drugs and generic drugs. The encoding method is [1,0] or [0,1].
[0080] Enterprise location: Assume that the enterprise location has three regions: Region 1, Region 2, and Region 3. The coding method is similar to that of the original research unit.
[0081] 2. Coding for therapeutic areas and indications.
[0082] Since these two key information may have a large number of categories and are unevenly distributed in different projects, the TF-IDF coding formula is used to calculate the coding value of each category.
[0083] Determine the number of categories: First, you need to determine the number of categories for key information such as therapeutic areas and indications. For example, a drug’s therapeutic areas may include multiple categories such as tumors, cardiovascular, and neurology; its indications may include multiple categories such as hypertension, diabetes, and cancer.
[0084] Arrange the sequence to be filled: Then, arrange these categories in a preset order to obtain a sequence to be filled. For example, the sequence to be filled in the therapeutic field may be [tumor, cardiovascular, neurological, ...]; the sequence to be filled in the indication may be [hypertension, diabetes, cancer, ...].
[0085] Calculate TF-IDF encoding value: Next, use the TF-IDF encoding formula to calculate the encoding value of each category in all projects. TF (term frequency) indicates the frequency of occurrence of a category in the current project; IDF (inverse document frequency) indicates the rarity of a category in all projects. By combining TF and IDF, we can get an encoding value that takes into account both the importance of the category in the current project and its rarity in all projects.
[0086] For example, suppose we have three documents: • Document 1: ["I", "love", "learning", "machine", "learning"] • Document 2: ["Me", "Like", "Programming"] • Document 3: ["Programming", "Very", "Interesting"] Step 1: Calculate TF Suppose we calculate the TF of the word "learning" in document 1: Step 2: Calculate IDF Suppose we calculate the IDF of "learning", assuming there are 3 documents in total, of which "learning" appears in document 1 and document 2.
[0087] Step 3: Calculate TF-IDF Finally, we can calculate the TF-IDF of "learning" in document 1.
[0088] Fill in the code value: Finally, fill the calculated code value of each category into the corresponding position of the sequence to be filled to obtain the code of the treatment area and indication.
[0089] 3. Specific Examples Suppose there is a drug development project with the following key information: Special Review: Fast Track Project Name: Project D (assumed to be the unique identifier of the fifth project) Original research unit: Unit B Highest stage: Phase III clinical trials and beyond Drug classification: Biological drugs Drug type: Innovative drug Company location: Region 2 Therapeutic Area: Oncology (assuming that the TF-IDF value of Oncology in all items is 0.8) Indication: Cancer (assuming that the TF-IDF encoding value of cancer among all items is 0.6) The coding process of this project is as follows: Using One-hot encoding: Special review: [1,0,0] Project name: [0,0,0,0,1] Original research unit: [0,1,0] Highest stage: [0,0,0,1] Drug classification: [0,1,0] Drug Type: [1,0] Company location: [0,1,0] Using TF-IDF encoding: Therapeutic Area: [0.8] (assuming only one dimension, tumor) Indication: [0.6] (assuming only one dimension, cancer) The above codes are concatenated in the preset order to obtain the coding vector of the project. For example, suppose we concatenate in the following order: special review, project name, original research unit, highest stage, drug classification, drug type, company location, therapeutic area, indication, then the coding vector of the project is: [1,0,0,0,0,0,0,1,0,0,0,0,1,0,0,0,0,1,0,0,0,0,0,0.8,0.6].
[0090] It should be noted that the coding values and order in the above examples are for illustration only and need to be adjusted according to specific projects and coding rules during the actual coding process.
[0091] Embodiment 3 In order to facilitate those skilled in the art to better understand the present method, a specific example is used for schematic illustration.
[0092] Step 1: Historical project information coding and processing First, we collected data from a series of historical drug development projects, each of which contained key information such as drug classification, drug type, company name, and company location. In order to uniformly process this information, we coded it according to the preset coding rules.
[0093] For example, for a historical project, the key information is: drug classification is "anti-tumor drug", drug type is "innovative drug", company name is "XX Pharmaceutical Company", and company location is "Shanghai". According to the coding rules, we selected the corresponding characters to represent this information, and combined them in the preset coding structure order to form a complete code. Finally, the generated code is verified to ensure that it meets the requirements of the coding rules, and the verified code is stored in the project coding database.
[0094] Step 2: Information coding and processing of the drug development project to be predicted Then, the same information encoding process was performed on a drug development project to be predicted. The key information of this project is: the drug classification is "cardiovascular drugs", the drug type is "generic drugs", the company name is "YY Pharmaceutical Company", and the company location is "Beijing". The encoding process was also performed according to the encoding rules, and the encoding vector of the project was obtained.
[0095] Step 3: Similarity matching and reference project selection Then, the cosine similarity algorithm was used to calculate the similarity between the coding vectors of the drug development project to be predicted (coding vector A) and the historical projects (coding vectors B1 to B7). Assuming that the coding vector of the project to be predicted is A and the coding vectors of the historical projects are B1 to B7, the cosine similarity between A and these vectors is calculated, and the results are as follows: Similarity between A and B1: 0.85 Similarity between A and B2: 0.78 Similarity between A and B3: 0.72 Similarity between A and B4: 0.68 Similarity between A and B5: 0.65 Similarity between A and B6: 0.55 Similarity between A and B7: 0.48 According to the similarity sorting, the top 5 items with the highest similarity (B1, B2, B3, B4, B5) are selected as reference items. In order to ensure the accuracy and reliability of the prediction, only the top N items with the highest similarity (N=5 in this case) are selected as references.
[0096] Step 4: Calculate the general average duration Next, based on the research and development stage to be predicted (for example, clinical trial phase I), calculate the average duration of each reference project in the corresponding research and development stage. Assuming that the duration of reference projects B1, B2, B3, B4, and B5 in clinical trial phase I is 18 months, 20 months, 19 months, 21 months, and 17 months, respectively, then the general average duration is (18+20+19+21+17) / 5=19 months.
[0097] Step 5: Determine the influencing factors and average duration deviation In order to more accurately predict the completion time of the project to be predicted, some influencing factors need to be considered. Based on historical project data, we conducted a quantitative analysis of various factors of drug development projects and found that the therapeutic field and the original research unit are the two factors that have the greatest impact on the development time. Therefore, these two factors are used as influencing factors of the project to be predicted, and the corresponding weight values are determined according to their degree of influence (for example, the weight of the therapeutic field is 0.6, and the weight of the original research unit is 0.4).
[0098] For the therapeutic area, we set screening criteria (for example, selecting historical projects in the same therapeutic area as the project to be predicted) and screened out projects that met the screening criteria from the reference projects. Then, we calculated the average duration of these projects in Phase I clinical trials and compared it with the universal average duration to obtain the average duration deviation in the therapeutic area. Similarly, we also calculated the average duration deviation of the original research unit.
[0099] For example, assuming that the average duration of Phase I clinical trials for projects with the same therapeutic area is 20 months, then the average duration deviation for the therapeutic area is 20-19 = 1 month. Similarly, assuming that the average duration of Phase I clinical trials for projects with the same original research unit is 18 months, then the average duration deviation for the original research unit is 18-19 = -1 month.
[0100] Step 6: Calculate the forecast completion date Finally, based on the general average duration and the average duration deviation of each influencing factor, the formula was used to calculate the predicted completion date of the drug development project to be predicted in Phase I of the clinical trial.
[0101] The formula is as follows: predicted completion time = current date + general average duration + average duration difference of each influencing factor × weight value.
[0102] Assuming that the current date is January 1, 2023, the predicted completion time is: January 1, 2023 + 19 months + 1×0.6 + (-1)×0.4 = September 1, 2024 (it should be noted that this is just a simple date addition. In actual use, it may be necessary to consider factors such as the number of days in the month and leap years for more accurate date calculations).
[0103] In summary, through the drug development time prediction method based on coding quantization and similarity matching, we can combine historical project data and influencing factors to more accurately predict the completion time of the project to be predicted, providing strong support for the progress management and resource allocation of drug development projects.
[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit the technical solution. Those skilled in the art should understand that those modifications or equivalent substitutions of the technical solution of the present invention that do not depart from the purpose and scope of the technical solution should be included in the scope of the claims of the present invention.
Claims
1. A drug development time prediction method based on coding quantization and similarity matching, characterized in that: The following steps are involved: S1. Obtain historical items, perform information encoding processing according to preset encoding rules, obtain corresponding encoding vectors and store them in the item encoding database; S2. According to the same coding rules, information encoding is performed on the drug development project to be predicted to obtain a corresponding coding vector; S3, according to the coding vector of the drug development project to be predicted, matching N historical projects with the highest coding vector similarity from the project coding database as reference projects; S4. Based on the R&D stage to be predicted, calculate the average duration of each reference project in the corresponding R&D stage, and record it as the general average duration; S5. Determine the influencing factors of the drug development project to be predicted; for each influencing factor, formulate screening conditions for it, and screen out reference projects that meet the screening conditions from the reference projects obtained in S3 as factor screening projects, and then calculate the average duration of the factor screening projects in the development stage to be predicted, and compare it with the general average duration to obtain the average duration deviation of the influencing factor; S6. Calculate the predicted completion date of the drug development project to be predicted in the research and development stage to be predicted based on the general average duration obtained in S4 and the average duration deviation of each influencing factor obtained in S5.
2. The method for predicting drug development time based on coding quantization and similarity matching according to claim 1, characterized in that: In S1, the preset coding rules include uniqueness, comprehensive coverage of key information, scalability, and readability.
3. The method for predicting drug development time based on coding quantization and similarity matching according to claim 2, characterized in that: The information encoding process includes: Obtain key information of the project; Encode each key information of the project according to a preset encoding rule to obtain the encoding of each key information; Combine the codes of each key information according to the preset coding structure sequence to form a complete code; The generated code is verified to ensure that it meets the requirements of the coding rules; and the verified code is used as the information coding processing result.
4. The drug development time prediction method based on coding quantization and similarity matching according to claim 3, characterized in that: Key project information includes therapeutic area, indication, special review, project name, original research unit, highest stage, drug classification, drug type and company location; When encoding the key information of the project, the one-hot coding method is used to encode special review, project name, original research unit, highest stage, drug classification, drug type and company location.
5. The method for predicting drug development time based on coding quantization and similarity matching according to claim 4, characterized in that: When coding the key information of the project, the process of coding the therapeutic areas and indications includes: First, determine the number of categories of this key information; Then, all categories of the key information are arranged in a preset order to obtain a sequence of the key information to be filled; Then, the coding values of each category of the key information are calculated according to the preset method; Finally, fill the code value of each category into the corresponding position of the sequence to be filled to obtain the code of the key information.
6. The method for predicting drug development time based on coding quantization and similarity matching according to claim 5, characterized in that: Use the TF-IDF coding formula to calculate the coding value for each category.
7. The method for predicting drug development time based on coding quantization and similarity matching according to claim 1, characterized in that: In S3, the cosine similarity algorithm is used to calculate the similarity of the encoding vectors between the drug development project to be predicted and the historical projects.
8. The method for predicting drug development time based on coding quantization and similarity matching according to claim 1, characterized in that: In S5, the process of determining the influencing factors of the drug development project to be predicted includes: based on historical project data, quantitatively analyzing various factors of the drug development project to determine their degree of influence on the development time; taking the M factors with the highest degree of influence as influencing factors, and determining the weight value of each influencing factor according to the degree of influence.
9. The method for predicting drug development time based on coding quantization and similarity matching according to claim 8, characterized in that: In S6, the predicted completion date of the drug development project to be predicted in the research and development stage to be predicted is calculated by the following formula: Predicted completion time = current date + general average duration + average duration difference of each influencing factor × weight value.
10. The method for predicting drug development time based on coding quantization and similarity matching according to claim 8, characterized in that: Influencing factors include therapeutic area and original research unit.