Kaiming Tao

Objectives: The ability of large language models (LLMs) to answer targeted scientific questions by synthesizing information from research articles remains an open research challenge. Methods: We evaluated the effects of fine-tuning and a question-specific prompting strategy to answer 16 pre-defined questions about HIV drug resistance studies, including whether viral genetic sequences were reported and the demographics and antiviral treatments of the individuals from whom sequences were obtained. For fine-tuning, we constructed an instruction set comprising 250 HIV drug resistance studies, with 16 questions per study and corresponding answers and explanations. For question-specific prompting, we developed a set of if-then rules tailored to each question. We compared the performance of three base models – GPT-4o-mini-2024-07-18 (GPT-4o), Meta Llama-3.1-70B-Instruct (Llama-3.1-70B), and Meta Llama-3.1-8B-Instruct (Llama-3.1-8B) – with their performance using fine-tuning, question specific prompting, and fine-tuning followed by question-specific prompting. Performance was assessed using accuracy, precision, recall, and F1 score, averaged over 150 held-out studies not used for fine-tuning. Comparisons were performed using Wilcoxon signed-rank tests. Results: Fine-tuning increased precision by 5% for GPT-4o, 16% for Llama-3.1-70B, and 8% for Llama-3.1-8B, although this increase reached statistical significance only for Llama-3.1-70B. Fine-tuning also significantly increased recall for GPT-4o by 11%. Question specific prompting increased recall for all three models (6% for GPT-4o, 7% for Llama-3.1-70B, and 18% for Llama-3.1-8B), with statistically significant improvements observed only for Llama-3.1-8B. Applying question specific prompting to each of the fine-tuned models did not yield additional improvements beyond fine-tuning alone. When pooled across the three models, fine-tuning was associated with a greater effect on precision than recall (OR = 4.35; p = 0.001; Fisher’s exact test), whereas question-specific prompting led to a greater effect on recall than on precision (OR= 7.09; p = 0.0001; Fisher’s exact test). Conclusions: In this domain-focused proof-of-concept study, fine-tuning and question-specific prompting each led to improvement in one or more metrics for each of the three models. Pooled analyses indicated that fine-tuning improved precision, whereas question specific prompting preferentially improved recall.

Introduction: Drug resistance testing may improve the management of people living with HIV (PLWH) in several scenarios in low- and middle-income countries (LMICs). To guide assay development, the WHO published a target product profile (TPP) outlining two priority use cases (scenarios) for genotypic resistance testing: (1) PLWH with confirmed virological failure (VF) on an integrase strand transfer inhibitor (INSTI)-based regimen, such as tenofovir (TFV) disoproxil fumarate, lamivudine (3TC), and dolutegravir (DTG) and (2) heavily treated PLWH, including infants and young children, with confirmed VF after receiving multiple regimens including a boosted protease inhibitor (PI). An additional potential scenario includes PLWH testing positive for HIV-1 while on pre-exposure prophylaxis (PrEP). Methods: To identify drug-resistance mutations (DRMs) most likely to influence clinical management of PLWH in each WHO TPP scenarios and to inform development of assays that detect individual DRMs and the interpretation of sequence-based assays, we reviewed prevalence and in vitro susceptibility data on HIV-1 DRMs in the Stanford HIV Drug Resistance Database associated with the nucleoside RT inhibitor (NRTI), nonnucleoside RT inhibitor (NNRTI), PI, and INSTI classes and the capsid inhibitor lenacapavir. Results: In the first scenario, the most informative NRTI DRMs were K65R and M184V/I; and the most informative INSTI DRMs were G118R, N155H, Q148H/K/R, and R263K. In the second scenario, a broader spectrum of DRMs is likely to be clinically relevant, including additional NRTI DRMs, the PI DRMs associated with reduced susceptibility to darunavir, and the NNRTI DRMs associated with reduced susceptibility to etravirine and doravirine. In PLWH testing positive for HIV-1 despite PrEP, the most informative NRTI and INSTI DRMs overlap with those in the first scenario, together with the capsid DRMs reported in persons experiencing VF while receiving lenacapavir. Conclusions: As global ART programs increasingly rely on INSTI-based regimens, and as the number of heavily treated individuals and difficult-to-treat pediatric cases grows, many LMICs have begun introducing HIV drug resistance testing for patient management. Although sequence-based assays provide the most comprehensive information for managing individual PLWH, assays that detect individual DRMs are also likely to be highly useful in the three WHO TPP scenarios.

Introduction: Few data are currently available on the nonnucleoside reverse transcriptase (RT) inhibitors (NNRTI) resistance mutations selected in persons living with HIV-1 (PLWH) who develop virological failure while receiving rilpivirine (RPV). Methods: We analyzed pooled HIV-1 RT genotypic data from 280 PLWH in the multicenter EuResist database and 115 PLWH in the Stanford HIV Drug Resistance Database (HIVDB) who received RPV as their only NNRTI. Results: Among the 395 PLWH receiving RPV, 180 (45.6%) had one or more NNRTI-associated DRMs. Overall, 44 NNRTI-associated DRMs were identified, including 26 that occurred in two or more PLWHs. Seven mutations had a prevalence ≥10% among the 180 PLWH with one or more NNRTI-associated DRM: E138K (32.2%), V90I (25.0%), K101E (17.8%), Y181C (17.2%), E138A (13.9%), H221Y (12.2%), and K103N (10.6%). Y181C was significantly more likely to co-occur with K101E, V179F, H221Y, and M230L. Ten novel non-polymorphic mutations at known NNRTI-associated mutation positions were also identified, usually in just one PLWH: L100F, V108A, T139I, P225S, M230V, Y232C, and T240A/I/M/S. Conclusions: Our analysis extends the spectrum of mutations emerging in PLWH receiving RPV. Additional phenotypic characterization of RPV-selected mutations is necessary to better understand their biological and possible clinical significance.

Background: With the approval of the HIV-1 capsid inhibitor, lenacapavir, capsid sequencing will be required for managing lenacapavir-experienced individuals with detectable viremia. Successful sequence interpretation will require examining new capsid sequences in the context of previously published sequence data. Methods: We analyzed published HIV-1 group M capsid sequences from 21,012 capsid-inhibitor naïve individuals to characterize amino acid variability at each position and influence of subtype and cytotoxic T lymphocyte (CTL) selection pressure. We determined the distributions of usual mutations, defined as amino acid differences from the group M consensus, with a prevalence ≥ 0.1%. Co-evolving mutations were identified using a phylogenetically-informed Bayesian graphical model method. Results: 162 (70.1%) positions had no usual mutations (45.9%) or only conservative usual mutations with a positive BLOSUM62 score (24.2%). Variability correlated independently with subtype-specific amino acid occurrence (Spearman rho = 0.83; p < 1 × 10⁻⁹) and the number of times positions were reported to contain an HLA-associated polymorphism, an indicator of CTL pressure (rho = 0.43; p = 0.0002). Conclusions: Knowing the distribution of usual capsid mutations is essential for sequence quality control. Comparing capsid sequences from lenacapavir-treated and lenacapavir-naïve individuals will enable the identification of additional mutations potentially associated with lenacapavir therapy.

Monoclonal antibodies (MAbs) targeting the SARS-CoV-2 spike protein are among the most effective measures for preventing and treating COVID-19. However, SARS-CoV-2 Omicron variants contain many mutations in their spike receptor-binding domains, the target of all authorized MAbs.

SARS-CoV-2 Omicron variants contain many mutations in its spike receptor binding domain, the target of all authorized monoclonal antibodies (mAbs). Determining the extent to which Omicron variants reduced mAb susceptibility is critical to preventing and treating COVID-19. We systematically reviewed PubMed and three preprint servers, last updated February 22, 2022, of the in vitro activity of authorized mAbs against the Omicron variants. Thirty-three studies were eligible including 33 containing Omicron BA.1 susceptibility data and five that also contained Omicron BA.2 susceptibility data. The first two authorized mAb combinations, bamlanivimab/etesevimab and casirivimab/imdevimab, were inactive against the Omicron BA.1 and BA.2 variants. In 24 studies, sotrovimab (third authorized mAb) displayed a median 4.1-fold (IQR: 2.4-7.6) reduced activity against Omicron BA.1 and, in four studies, a median 26-fold (IQR:16-35) reduced activity against Omicron BA.2. In 18 studies, cilgavimab and tixagevimab independently displayed median reductions in activity of >300-fold against Omicron BA.1, while in ten studies, the cilgavimab/tixagevimab combination (fourth authorized mAb preparation) displayed a median 63-fold (IQR: 26-145) reduced activity against Omicron BA.1. In two studies, cilgavimab was approximately 100-fold more susceptible to BA.2 than to BA.1. In two studies, bebtelovimab, the most recently authorized mAb, was fully active against both the Omicron variants. Disparate results between assays were common as evidenced by a median 42-fold range (IQR: 25-625) in IC50 between assays for the eight authorized individual mAbs and three authorized mAb combinations. Highly disparate results between published assays indicates a need for improved mAb susceptibility test standardization or inter-assay calibration.

As novel SARS-CoV-2 variants with different patterns of spike protein mutations have emerged, the susceptibility of these variants to neutralization by antibodies has been rapidly assessed. However, neutralization data are generated using different approaches and are scattered across different publications making it difficult for these data to be located and synthesized. The Stanford Coronavirus Resistance Database (CoV-RDB; https://covdb.stanford.edu) is designed to house comprehensively curated published data on the neutralizing susceptibility of SARS-CoV-2 variants and spike mutations to monoclonal antibodies (mAbs), convalescent plasma (CP), and vaccinee plasma (VP). As of December 31, 2021, CoV-RDB encompassed 257 publications including 91 (35%) containing 9,070 neutralizing mAb susceptibility results, 131 (51%) containing 16,773 neutralizing CP susceptibility results, and 178 (69%) containing 33,540 neutralizing VP results. The database also records which spike mutations are selected during in vitro passage of SARS-CoV-2 in the presence of mAbs and which emerge in persons receiving mAbs as treatment. The CoV-RDB interface interactively displays neutralizing susceptibility data at different levels of granularity by filtering and/or aggregating query results according to one or more experimental conditions. The CoV-RDB website provides a companion sequence analysis program that outputs information about mutations present in a submitted sequence and that also assists users in determining the appropriate mutation-detection thresholds for identifying non-consensus amino acids. The most recent data underlying the CoV-RDB can be downloaded in its entirety from a GitHub repository in a documented machine-readable format.

The development of effective antiviral therapy for COVID-19 is critical for those awaiting vaccination, as well as for those who do not respond robustly to vaccination. This review summarizes 1 year of progress in the race to develop antiviral therapies for COVID-19, including research spanning preclinical and clinical drug development efforts, with an emphasis on antiviral compounds that are in clinical development or that are high priorities for clinical development.

As novel SARS-CoV-2 variants with different patterns of spike mutations have emerged, the susceptibility of these variants to neutralization by antibodies has been rapidly assessed. However, neutralization data are generated using different approaches and are scattered across different publications making it difficult for these data to be located and synthesized. The Stanford Coronavirus Resistance Database (CoV-RDB; https://covdb.stanford.edu ) is designed to house comprehensively curated published data on the neutralizing susceptibility of SARS-CoV-2 variants and spike mutations to monoclonal antibodies (mAbs), convalescent plasma (CP), and vaccinee plasma (VP). As of October 2021, CoV-RDB contains 186 publications including 64 (34%) containing 7,328 neutralizing mAb susceptibility results, 96 (52%) containing 11,390 neutralizing CP susceptibility results, and 125 (68%) containing 20,872 neutralizing VP results. The database also records which spike mutations are selected during in vitro passage of SARS-CoV-2 in the presence of mAbs and which emerge in persons receiving mAbs as treatment. The CoV-RDB interface interactively displays neutralizing susceptibility data at different levels of granularity by filtering and/or aggregating query results according to one or more experimental conditions. The CoV-RDB website provides a companion sequence analysis program that outputs information about mutations present in a submitted sequence and that also assists users in determining the appropriate mutation-detection thresholds for identifying non-consensus amino acids. The most recent data underlying the CoV-RDB can be downloaded in its entirety from a Github repository in a documented machine-readable format.